Corinne Brusque
(INRETS, France)
Proceedings of
European Conference
on Human Centred Design for
Intelligent Transport Systems
April 3rd-4th 2008
Lyon, France
HUMANIST publications – Lyon, 2008
HUMANIST C/O INRETS - 25 avenue François Mitterrand - Case 24 69 675 BRON cedex - FRANCE
ISBN: 978-2-9531712-0-4
All rights reserved. No part of this publication may be reproduced, stored or
transmitted in any form or by any means, without the prior written permission of
the authors.
Free specimen copies are available on request:
PREFACE
The widespread distribution of in-vehicle driver information systems on the
market and the emergence of advanced driver assistance systems are
profoundly changing road transport and more specifically driver environment
and driving activity. Through Intelligent Transport Systems, a range of services
are indeed offered to the driver with the objective of facilitating the driving task
and improving travel safety. Despite the potential benefits on road safety, these
developments raise numerous questions about their relevance, usability and
acceptability for drivers and their impact on drivers’ behaviour and attitudes. All
this contributes to the necessity of encouraging a Human Centred Design
approach, in which ITS are designed according to driver needs
That is the reason why the HUMANIST Network of Excellence decided to
organize a conference that would allow the community of Human Factors
researchers to gather in order to present and discuss current works on various
aspects of Human Centred Design for Intelligent Transport Systems.
The first European conference on Human Centred Design for Intelligent
Transport Systems held in Lyon on 3rd & 4th April 2008 was organized by
INRETS, the French National Institute for Transport and Safety Research and
by ERT, Europe Recherche Transport, with the support of Jean-Pierre
Médevielle, the Coordinator of the HUMANIST Network of Excellence.
This volume presents the conference edited proceedings. Thirty three peerreviewed papers are included and presented in six sessions: “Drivers’
distraction due to ITS use”, “Tools and methodologies for safety and usability”,
“Modelling of drivers’ behaviour for ITS design”, “Tools and methodologies for
ITS design and drivers awareness”, “Diversity and specificity of road user
groups” and “Drivers’ acceptance of assistance functions”. The volume provides
an extensive overview of the current developments and trends in Human
Centred Design for Intelligent Transport Systems.
The conference scientific committee included Corinne Brusque (INRETS,
France), Martin Baumann (DLR, Germany), Guy Boy (EURISCO, France),
Emilio Davila-Gonzalez (DG-INFSo, European Commission), John Golias
(NTUA, Greece), Josef F. Krems (Chemnitz TU, Germany), José Manuel
Menendez (UPM, Spain), Merja Penttinen (Nokia, US), Pirkko Rama (VTT,
Finland), Ralf Risser (FACTUM, Austria), Alan Stevens (TRL, UK), Divera Twisk
(SWOV, The Netherlands), Mark Vollrath (Braunschweig TU, Germany).
iii
I wish to express my gratitude towards the scientific committee members and
the following researchers who actively participated in peer-review process:
Angelos Amditis (ICCS, Greece), Angelos Bekiaris (CERTH/HIT, Greece),
Thierry Bellet (INRETS, France), José Carvalhais (UTL, Portugal), André
Chapon (INRETS, France), Michael Falkenstein ((IfADo,Germany), Alexandra
Fort (INRETS, France), Christhard Gelau (BASt, Germany), Richard van der
Horst (TNO, The Netherlands), Roberto Montanari (UNIMORE, Italy), José
Pardilllo Mayora (UPM, Spain), Michael Regan (INRETS / MUARC, Australia),
Karel Schmeidler (CDV, Czech Republic), Anabela Simoes (ISEC, Portugal),
Mark Tant (IBSR, Belgium) and Truls Vaa (TOI, Norway).
The scientific committee conveys its thanks to Soizick Chambard, Céline Goupil
and Nicole Teillac of INRETS who handled the all conference organization.
We are also very grateful to the following bodies for their support:
¾
¾
¾
¾
European Commission
DG INFSo Information Society and Media Directorate General
Région Rhône-Alpes
Fondation Rhône-Alpes Futur
Corinne Brusque
INRETS, France
iv
CONTENTS
Preface ................................................................................................................iii
Contents .............................................................................................................. v
Session 1: Drivers’ distraction due to ITS use
The effect of cognitive tasks on predicting events in traffic ............................. 3
Martin R.K. Baumann (DLR / BASt, Germany), Tibor Petzoldt (Chemnitz
University of Technology, Germany), Chantal Groenewoud (Utrecht
University, The Netherlands), Jeroen Hogema (TNO, The Netherlands),
Josef F. Krems (Chemnitz University of Technology, Germany)
Neural basis for some cognitive risk of using mobile phones during
driving ............................................................................................................ 13
Takashi Hamada (AIST, Japan)
Effects of visual search task complexity on lane change task
performance................................................................................................... 23
Ellen S. Wilschut (IfADo, Germany), Gerhard Rinkenauer (IfADo,
Germany), Karel A. Brookhuis (Groningen University, The Netherlands),
Michael Falkenstein (IfADo, Germany)
Effects of simultaneous multi-modal warnings and traffic information on driver
behaviour ....................................................................................................... 33
Clemens Kaufmann (FACTUM OHG, Austria), Ralf Risser (FACTUM OHG,
Austria), Arjan Geven (CURE, Austria), Reinhard Sefelin (CURE, Austria)
Naturalistic driving observations to investigate distraction exposure and IVIS
patterns of use: interests and constraints of the approach............................ 43
Arnaud Bonnard, Corinne Brusque (INRETS, France)
Session 2: Tools and methodologies for safety and usability
Development of innovative methodologies to evaluate its safety and usability:
HUMANIST TF E ........................................................................................... 55
Sally Cotter (TRL, United Kingdom), Alan Stevens (TRL, United Kingdom),
Anke Popken (Chemnitz University of Technology, Germany), Christhard
Gelau (BASt, Germany)
Evaluating driver mental workload using the driving activity load index
(DALI) ............................................................................................................ 67
Annie Pauzié (INRETS, France)
Comparison of subjective workload ratings and performance measures of a
reference IVIS task ........................................................................................ 79
Tony Wynn (ESRI, United Kingdom), John, H. Richardson (ESRI, United
Kingdom)
v
Impairment of Lane change performance due to distraction: effect of
experimental contexts .................................................................................... 89
Marie-Pierre Bruyas, Corinne Brusque, Adrien Auriault, Hélène Tattegrain,
Isabelle Aillerie, Max Duraz (INRETS, France)
Connecting black box data and driving behaviour observation for better
understanding of driving behaviour.............................................................. 101
Clemens Kaufmann, Christine Turetschek (FACTUM OHG, Austria)
Session 3: Modelling of drivers’ behaviour for ITS design
Modelling driver behaviour in order to infer the intention to change lanes .. 113
Matthias J. Henning (Chemnitz University of Technology, Germany),
Olivier Georgeon (INRETS, France), Tony Wynn (SERI, United Kingdom),
Josef F. Krems (Chemnitz University of Technology, Germany)
Analysis and modelling of driver preparatory behaviour before turning at
intersections................................................................................................. 121
Toshihisa Sato, Motoyuki Akamatsu (AIST, Japan)
Influence of unexpected events on driving behaviour at different hierarchical
levels: a driving simulator experiment.......................................................... 131
T. W. Schaap (Twente University, The Netherlands), A. R. A. van der Horst
(TNO, The Netherlands), B. van Arem (Twente University, The
Netherlands)
Subjective strain estimation depending on driving manoeuvres and traffic
situation........................................................................................................ 141
Caroline Schießl (DLR, Germany)
Identifying user strategies for interaction with in–vehicle information systems
while driving in a simulator........................................................................... 153
Nadja Rauch, Barbara Gradenegger,
Wuerzburg University, Germany)
Hans-Peter
Krüger
(IZVW,
Parameters related to modelling intelligent speed adaptation systems with the
employment of a microscopic traffic model ................................................. 163
Ioanna Spyropoulou, Matthew G. Karlaftis (National Technical University
of Athens, Greece)
Session 4: Tools and methodologies for ITS design and drivers awareness
A service-oriented system architecture for the human centered design of
intelligent transportation systems ................................................................ 175
Jan Gačnik, Oliver Häger, Marco Hannibal (DLR, Germany)
Analysis and modelling methods for the design and evaluation of an ecodriving system .............................................................................................. 183
Jérôme Barbé, Guy Boy (EURISCO International, France)
vi
An analysis of the requirements of driver assistance systems – when and
why does the driver like to have assistance and how can this assistance be
designed? .................................................................................................... 193
Julia Werneke (DLR, Germany), Astrid Kassner (DLR, Germany), Mark
Vollrath (Braunschweig University of Technology, Germany)
A Human factor-based approach for the effective use of driving simulators
and e-learning Tools in driver training and education ................................. 205
Jose M Pardillo Mayora (Technical University of Madrid, Spain)
Comparative evaluation of training methods in improving drivers’
understanding about the functionalities and potential limitations
of ADAS ....................................................................................................... 217
Villy Portouli (CERTH/HIT, Greece), Evangelos Bekiaris (CERTH/HIT,
Greece), Sofie Boets (IBSR, Belgium), Per Henriksson (VIT, Sweden)
Session 5: Diversity and specificity of road user groups
Applying the response code of practice for evaluation of driver assistance
systems: how the age of the driver influences perception of steering torque
signals.......................................................................................................... 229
Gerrit Schmidt (Volkswagen AG, Germany)
Effect of additional tasks in driving performance: comparison among three
groups of drivers .......................................................................................... 239
Marta Pereira (UTL / FMH, Portugal), Houda Hamama (INRETS, France),
Marie-Pierre Bruyas (INRETS, France), Anabela Simões (ISEC, Portugal)
What tools are needed to develop safe and joyful cycling for senior
citizens......................................................................................................... 253
Lars Leden (VTT, Finland)
Preliminary results of a study about the influence of an active accelerator
pedal on novice and young drivers in Austria and the Czech Republic ...... 265
Christine Turetschek (FACTUM OHG, Austria)
Cognitive activity modelling: a case study of lane change schemas and
sensation seeking ........................................................................................ 277
Tony Wynn (ESRI, United Kingdom), John H. Richardson (ESRI, United
Kingdom), Olivier Georgeon (INRETS, France), Thierry Bellet (INRETS,
France), Matthias J. Henning (Chemnitz University of Technology,
Germany), Josef F. Krems (Chemnitz University of Technology, Germany)
Risk awareness analysis: a comparison between car drivers and
motorcyclists ................................................................................................ 287
Aurélie Banet, Thierry Bellet (INRETS, France)
Session 6: Drivers’ acceptance of assistance functions
vii
Drivers’ reliance on lane keeping assistance systems: Effects of different
Levels of assistance .................................................................................... 301
Anke Popken (Chemnitz University of Technology, Germany), Lena
Nilsson (VTI, Sweden), Josef
F. Krems (Chemnitz University of
Technology, Germany)
User acceptance and effectiveness of warning and motor priming assistance
devices in car driving ................................................................................... 311
M. El Jaafari (Renault, France), J-F. Forzy (Renault, France), J. Navarro
(IRCCyN, France), F. Mars (IRCCyN, France), J-M. Hoc (IRCCyN, France)
ReflektAS – Usage of reflexive driver reactions in ADAS-development...... 321
Ingmar Ehrenpfordt (DLR, Germany), Anke Kullack (Braunschweig
University of Technology, Germany), Frank Eggert (Braunschweig
University of Technology, Germany)
Meet the driver needs by matching assistance functions and task
demands ...................................................................................................... 327
Astrid Kassner (DLR, Germany)
Perceptions of risk and control understanding acceptance of advanced driver
assistance systems...................................................................................... 335
Somya Joshi (ICCS, Greece), Angelos Amditis (ICCS, Greece), Thierry
Bellet (INRETS, France), Vanessa Bodard (INRETS, France)
The effect of ISA in relation to speed limit credibility ................................... 347
Nicole van Nes, Ingrid van Schagen (SWOV, The Netherlands)
Index of authors ............................................................................................... 357
viii
SESSION 1:
DRIVERS’ DISTRACTION
DUE TO ITS USE
1
Drivers’ distraction due to ITS use
THE EFFECT OF COGNITIVE TASKS ON
PREDICTING EVENTS IN TRAFFIC
Martin R.K. Baumann (DLR / BASt, Germany),
Tibor Petzoldt (Chemnitz University of Technology, Germany),
Chantal Groenewoud (Utrecht University, The Netherlands),
Jeroen Hogema (TNO, The Netherlands),
Josef F. Krems (Chemnitz University of Technology, Germany)
martin.baumann@dlr.de
ABSTRACT: Numerous studies demonstrate the negative effects of cognitively
loading secondary tasks on driving performance. We assume that this effect is
caused by interference between these secondary tasks and central executive
functions of working memory that serve to keep the driver’s situation model of
the current traffic situation updated. In this experiment 48 drivers had to drive in
a high fidelity driving simulator on a rural road while performing no secondary
task, or a working memory task (auditive monitoring) that should not interfere
with situation awareness, or a working memory task (memory updating) that
should interfere with the comprehension and prediction function of situation
awareness. While driving, participants had to react to events that were either
announced by a warning signal or not. We hypothesized that participants will
benefit least from the warning signal when they had to perform the memory
updating task. The results generally support this hypothesis indicating that
central executive functions of working memory are highly involved in situation
awareness processes.
1 Introduction
There are numerous studies demonstrating the negative effects of cognitive
tasks on driving performance (e.g., [1-9]). These studies generally find an
increase in response latencies of drivers performing cognitively loading tasks, a
decrement in lane keeping performance, or a loss of situation awareness. But
these studies seldom link their findings to underlying cognitive processes. The
aim of this experiment is to examine whether the effect of cognitively loading
tasks on driving behaviour can be partly attributed to interference between the
loading task and situation awareness processes. According to [10] situation
awareness serves three functions: the perception of elements of a situation, the
comprehension of these elements and their relation to the situation as a whole,
and the prediction of the future development of the situation. Especially the
comprehension and the prediction function draw on working memory resources.
To accomplish these functions perceived information has to be associated with
information stored in long-term memory to construct a knowledge network that
represents the meaning of the current situation, the situation model. We will use
this term synonymously to situation awareness. For this construction process
knowledge has to be retrieved from long-term memory in order to be available
for the associative processes. [11] assume that the prediction function is
integrated within this comprehension process as in most cases a given situation
is not only connected to knowledge that determines its meaning but also to
3
Human Centred Design for Intelligent Transport Systems
expectations about its future development. These expectations are also
retrieved and become part of the situation model. In non-routine situations
additional attention-demanding processes are necessary to make predictions
about the further development of the situation. In this case information has to be
kept in working memory to be available for these processes.
Imposing cognitive load on the driver withdraws resources necessary for the
comprehension of the current situation and the prediction of future development
of the situation. The looked-but-did-not-see phenomenon [12] is an example of
the interference of cognitive load with situation awareness. In this case the
cognitive load by an additional task leads to an incomplete comprehension of
one or more situation elements. This then might lead to an inappropriate action
selection of the driver.
According to [13, 14] working memory is not a single structure but consists of
different parts: a phonological buffer with an articulatory loop, a visuo-spatial
buffer, and the central executive. Recent studies [14-16] indicate that the central
executive serves different functions that seem to be highly relevant for the
comprehension and projection processes of situation awareness as described
above. One of these central executive functions is the retrieval of information
from memory, a process essential in the comprehension process of situation
awareness. Another is the maintenance of information in a non-modality specific
buffer, the episodic buffer, and the control of working memory contents. We
assume that the effect of cognitive loading tasks on driving is at least in part due
to the interference of these tasks with these central executive processes. The
specific aim of this experiment is to test whether specific kinds of cognitive load
that are designed to interfere with specific central executive processes interfere
especially with the prediction function of situation awareness.
Therefore the participants in this experiment drove through a scenario that
contained both predictable and non-predictable events. These events were
designed to be exactly equivalent besides that in the predictable version a
warning sign warned the driver of the upcoming event. The reaction to the event
when the participant was warned was compared to the reaction when the driver
was not warned.
While driving the participants had to perform (i) an auditory monitoring task that
should not load on the comprehension functions of the central executive, or (ii)
a running memory task that should heavily load on those central executive
functions that are involved in the comprehension and prediction function of
situation awareness, or (iii) no secondary task. In the monitoring task
participants had to react as fast as possible to an auditory signal that was
presented with either after a long or a short time interval after the previous
signal. By using only two randomly presented interstimulus intervals this tasks
induces a strong tendency for rhythmic responding leading to errors, mainly too
early responses. To avoid these errors one has to constantly suppress rhythmic
tapping. According to [17] this task should tap the monitoring function of the
central executive. But as this function should not be strongly involved in the
construction process of situation awareness this task should interfere less with
the construction of a situation model. Therefore the monitoring task should
interfere less with the prediction of events in traffic than the running memory
task.
4
Drivers’ distraction due to ITS use
In the running memory task [18] participants are presented with a constant
stream of items and they have to remember the last items, for example the last
three items. As the participants do not know when the stream ends and
therefore do not know when they are asked to recall the last presented items,
this task requires that the set of items kept in working memory is constantly
updated. That is, each time a new item is presented it has to be encoded in
working memory and the “oldest” item has to be removed from working
memory. Performing this task directly involves those central executive functions
that control working memory content, i.e. those functions that should also be
highly involved in maintaining and updating a proper situation model. Therefore,
this task should interfere with situation awareness processes and especially
with the prediction function of situation awareness. We assume therefore that
this updating of working memory is highly interfering with the comprehension
and the projection function of situation awareness.
To summarize, we assumed that participants driving the scenario without
performing a secondary task should clearly benefit from the warning signs in the
predictable events. The benefit from warning signs should be reduced when
participants had to perform an additional task while driving. And the reduction of
this benefit should be greater when participants had to perform the running
memory task than when they had to perform the monitoring task, as the
memory task should interfere more with the comprehension and prediction
function of situation awareness than the monitoring task.
2 Method
2.1
Participants
48 participants took part in this experiment. Their age ranged from 21 to 58 with
a mean of 36.9 years (SD = 12.1). Of these participants 29 were male. All
participants possessed a valid driving licence at least for one year and drove at
least 10000 km per year.
2.2
Driving scenario
The experiment was run in the high fidelity driving simulator of TNO in
Soesterberg, Netherlands. The driving scenario consisted of driving on a rural
road with an approximate speed of 80 km/h. Each drive took about 20 min.
During a drive each participant encountered four critical events. In all of these
events the participant’s lane was blocked by an obstacle, for example by a
construction site. In two of these events the driver was given information to
predict the obstacle, for example a warning sign next to the road; in the other
two the driver did not receive such warning information. Each participant drove
the scenario only once.
5
Human Centred Design for Intelligent Transport Systems
Fig.1. Sketch of the basic layout of the critical events; left side non-predictable
obstacle on the lane as no warning sign was presented, right predictable
obstacle with warning sign (marked by triangle)
2.3
Secondary tasks
In each secondary task condition participants had to perform several trials of the
respective task. The trials lasted between 20 and 40 sec, and were followed by
a no secondary task phase (where participants only had to drive) also lasting
between 20 and 40 sec. Start and the end of a secondary task trial were
triggered when the participant passed a certain position on the road. Therefore
it was guaranteed that the participants in the secondary task conditions
encountered the critical events while performing the respective secondary task.
2.3.1 The monitoring task
In the monitoring task participants had to react as fast as possible to an
acoustical signal consisting of a short, clearly audible sound. As response
device a finger switch was used that was applied to the index finger of the
participant’s dominant hand. The time interval between two successive signals
was either 1 or 2 sec, randomly chosen. We measured the participants’
response times and the numbers of errors, predominantly early responses.
2.3.2 The running memory task
In the running memory task participants were presented an audio stream of 13
possible letters, presented with a frequency of 1 letter per 2 sec. The
participants’ task was to repeat the current last three letters each time a new
letter was presented. For example, assume the letters “S”, “P”, and “Q” were
already presented and the next letter was “G”, the participant had to repeat
loudly “P Q G”, after presentation of “G”, and after the next letter “M”, the
participant had to repeat “Q G M”, and so on. After a variable amount of time
(20 to 40 sec) an acoustical signal was presented to inform the participant about
the end of the current trial. After that the last repeated triple of letters was taken
as the participant’s response in this trial. This response was taken to measure
the participant’s accuracy in this task.
2.4
Design
In this experiment two independent variables were manipulated: the type of
secondary task (no secondary, monitoring task, running memory task) and the
predictability of the critical event (predictable, non-predictable). The secondary
task factor was manipulated as between-subjects factor. Therefore the whole
6
Drivers’ distraction due to ITS use
sample of 48 participants was divided into three groups with 16 participants
each. The predictability factor was manipulated as within-subjects factor. Each
of the participants encountered both predictable and non-predictable events
during his/her drive. This results in a 3 (secondary task) x 2 (predictability)
mixed factorial design.
As dependent measures we used different aspects of the driving performance
to characterize participants’ reaction to the critical events, such as Time to
Collision at the time of response begin or maximum brake pressure after the
obstacle became visible. Also, driving performance in phases without critical
events was recorded to allow for the assessment of the effects of the secondary
task performance on normal driving. Additionally, the participants were asked to
rate their workload when performing each task while driving using the Rating
Scale for Mental Effort (RSME, [19]). We also measured the participants’
secondary task performance and eye movement behaviour but describing the
analysis of these measures is beyond the scope of this paper.
3 Results and Discussion
We will only present an overview of some critical results of this experiment.
These results focus on the driver’s reaction to the obstacle on the lane after it
became visible to the driver (see Table 1). After this moment there should be a
clear difference in drivers’ reaction between drivers that were warned of the
obstacle before and were able to correctly comprehend and integrate its
meaning into their situation model compared to drivers that were either not
warned or not able to fully comprehend the warning signal and therefore were
not able to predict the obstacle on the lane. Therefore the difference in the
respective measures of drivers’ reactions to events with warning signal and to
the corresponding events without warning signal reflect how effective the
participants in the different secondary task conditions could integrate the
warning signal into their situation model and prepare themselves for the
upcoming obstacle. According to our hypothesis we assume that this difference
should be greatest in the no secondary task condition and smallest in the
memory updating condition.
Table 1. Mean values of various performance measures
TTC at first throttle release after
roadblock is visible in sec
Speed at first throttle release after
roadblock is visible in kph
no sign
sign
no sign
sign
no
secondary
task
3.69
5.26
70.23
57.62
monitoring
task
3.75
4.55
68.77
62.86
memory
updating
task
3.66
4.24
69.21
64.45
The first measure is the Time to Collision (TTC) at the moment the driver
released the throttle to brake after passing the location where the obstacle first
became visible. As participants being prepared to the obstacle should have
already reduced their speed after seeing the warning signal and should brake
earlier than participants that did not comprehend the warning signal fully, TTC
for prepared participants should be larger than for unprepared. And as stated
above this difference should be greatest in the no secondary task condition and
7
Human Centred Design for Intelligent Transport Systems
lowest in the updating memory condition. As shown in the first row in Table 1
the results confirm this prediction. Whereas the difference between the
predictable and the non-predictable obstacle is 1.6 sec in the no secondary task
condition, it is 0.8 sec in the monitoring task condition and 0.6 sec in the
memory updating condition. This picture is confirmed in a 3 (secondary task) x 2
(predictability) mixed ANOVA. TTC is significantly greater for predictable
events, F(1, 45) = 40.78, p < .001. And very important the interaction between
secondary task condition and predictability is significant, F(2, 45) = 3.86, p =
.028, reflecting the reduction in the difference between predictable and nonpredictable events from no secondary task to monitoring task to running
memory task condition. The main effect of secondary task condition did not
reach significance, F(2, 45) = 1.95, p = .15.
The same picture emerges when one looks at the speed of the participants
when they release the throttle to brake in front of the obstacle. First, there is not
much of a difference in speed between the no secondary task condition and the
two secondary task conditions in case of a non-predictable obstacle, indicating
the validity of the experimental design. But there is a clear difference in speed
between these conditions in case of a predictable obstacle. Participants in the
secondary task conditions were driving faster when they started to brake than
participants in the no secondary task condition, indicating that these participants
were less prepared to the obstacle despite the warning sign. Looking at the
speed difference between predictable and non-predictable obstacles the same
pattern can be found as for TTC. The greatest difference in speed could be
found in the no secondary task condition, 12.6 km/h, a medium difference in the
monitoring condition, 5.9 km/h, and the smallest in the memory updating
condition, 4.8 km/h. Again this indicates that participants in the memory
updating condition had the greatest difficulties to comprehend, integrate, and
react to the warning sign. A 3 (secondary task) x 2 (predictability) mixed
ANOVA revealed a significant main effect of predictability, F(1, 45) = 43.63, p <
.001, indicating that speed was significantly higher in non-predictable events at
the time of the begin of the brake reaction. As for TTC the interaction between
secondary task condition and predictability for speed as dependent measure
was significant, F(2, 45) = 4.35, p = .019, confirming the reduced benefit from
the warning signal from no secondary task to monitoring task to running
memory task condition.
A basic problem with this kind of analysis is that the interpretation of the effects
of the secondary tasks on driving performance as being due to structural
differences between the secondary tasks rests on the precondition that task
difficulty can be excluded as alternative cause. More specific, the greater
interference between the memory updating task and the construction of the
situation model should not be a result of a memory updating task that is simply
more difficult than the monitoring task. Instead, the greater interference should
be due to the different working memory functions both tasks involve. This
argument cannot be made straight forward and easily because of the necessary
differences between both tasks, especially with regard to the different
performance measures – number of trials with correctly updated memory set for
the memory updating task vs. number of wrong taps in the monitoring task. But
there are some indications that support the interpretation of the results in terms
of the structural task differences. First, we analysed lane keeping performance
8
Drivers’ distraction due to ITS use
in curves. Previous studies demonstrated the sensitivity of this driving
performance aspect to driver’s cognitive load [2, 5]. As the driving task is
especially difficult in curves, driving performance should be especially sensitive
to differences in task difficulty between the secondary tasks. We found no
differences between the tasks both with regard to standard deviation of lane
position and number of lane exceedances. A second indication that both tasks
were of comparable task difficulty stems from the participants’ RSME ratings of
workload when performing each task while driving. There was no difference in
the ratings between the tasks.
These results cannot completely exclude the alternative interpretation that the
greater interference between the memory updating task and the situation model
construction is due to the greater difficulty of the updating task. But these
results support the interpretation that structural differences between the tasks
are the reason for the differential effects on situation awareness.
4 Conclusions
The aim of this experiment was to test one hypothesis about the cognitive
foundation of negative effects of cognitively distracting tasks on driving
performance. We assumed that such tasks load on working memory and
especially on central executive functions of working memory that are also highly
involved in maintaining and updating a proper situation model of the current
traffic situation. Such functions are the control of working memory content,
involving updating the content and removing irrelevant information from working
memory, and the retrieval of information from long-term memory to comprehend
encoded information in working memory. These functions are necessary to
keep the situation model current and to integrate all relevant implications of
observed situation elements into the situation model. If additional tasks interfere
with these functions as they also require them then the risk is that some
elements of the traffic situation are not fully comprehended and their
implications are not integrated into the situation model. This should clearly
impair the prediction function of situation awareness, that is cognitively
distracted drivers should be less able to predict the future development of the
current traffic situation and therefore should be less prepared to these
developments.
We tested this prediction in a driving simulator experiment where participants
encountered predictable and non-predictable events while driving and
performing secondary tasks that specifically tapped relevant central executive
functions of working memory. The results confirm our hypothesis that the
central executive function of controlling working memory content is highly
involved in the construction of situation awareness. Interfering with this function
by a secondary task leads to a detrimental effect in the driver’s ability to predict
the future development of a traffic situation. This results help to clarify the
cognitive mechanisms that underly situation awareness.
9
Human Centred Design for Intelligent Transport Systems
5 Acknowledgments
This study was funded by the Network of Excellence “Humanist” (Human
centred design for information society technologies). We would like to thank
Ingmar Stel from TNO for programming the driving simulation scenario,
preprocessing the data according to our changing requirements and supporting
this study with helpful hints. We also would like to thank Marika Hoedemaeker
and Wiel Janssen from TNO for their help in designing and conducting this
experiment.
6 References
[1]
[2]
[3]
[4]
[5]
[6]
[7]
[8]
[9]
[10]
[11]
[12]
[13]
McKnight AJ, McKnight AS: 'The effect of cellular phone use upon driver
attention'. Accident Analysis & Prevention, 1993, 25, (3), pp. 259-265
Engström J, Johansson E, Östlund J: 'Effects of visual and cognitive load
in real and simulated motorway driving'. Transportation Research Part F,
2005, 8, (2), pp. 97-120
Hancock PA, Lesch M, Simmons L: 'The distraction effects of phone use
during a crucial driving maneuver'. Accident Analysis & Prevention, 2003,
35, (4), pp. 501-514
Harbluk JL, Noy YI, Eizenman M: 'The impact of cognitive distraction on
driver visual behavior and vehicle control', Transport Canada, 2002, Report
No.: TP 13889 E.
Jamson AH, Merat N: 'Surrogate in-vehicle information systems and driver
behaviour: Effects of visual and cognitive load in simulated rural driving'.
Transportation Research Part F, 2005, 8, (2), pp. 79-96
Nunes L, Recarte MA: 'Cognitive demands of hands-free-phone
conversation while driving'. Transportation Research Part F: Traffic
Psychology and Behaviour, 2002, 5, (2), pp. 133-144
Salvucci DD, Macuga KL: 'Predicting the effects of cellular-phone dialling
on driver performance'. Cognitive Systems Research, 2002, 3, (1-4), pp.
95-102
Strayer DL, Johnston WA: 'Driven to distraction: Dual tasks studies of
simulated driving and conversing on a cellular telephone'. Psychological
Science, 2001, 12, (6), pp. 462-466
Brown ID, Tickner AH, Simmonds DC: 'Interference between concurrent
tasks of driving and telephoning'. Journal of Applied Psychology, 1969, 53,
(5), pp. 419-424
Endsley MR: 'Toward a theory of situation awareness in dynamic systems'.
Human Factors, 1995, 37, (1), pp. 32-64
Baumann M, Krems JF: 'Situation Awareness and Driving: A Cognitive
Model'. in: Cacciabue PC, (Ed.): 'Modelling Driver Behaviour in Automotive
Environments' (Springer, 2007), pp. 253-265
Brown ID: 'Review of the "Looked but Failed to See" Accident Causation
Factor'. London: Department for Transport; 2005. Report No.: 60.
Baddeley A: 'The episodic buffer: A new component of working memory?'
Trends in Cognitive Sciences, 2000, 4, (11), pp. 417-423
10
Drivers’ distraction due to ITS use
[14] Baddeley AD: 'Working Memory' (Clarendon Press, 1986)
[15] Baddeley A, Emslie H, Kolodny J, Duncan J: 'Random generation and the
executive control of working memory'. Quarterly Journal of Experimental
Psychology: Human Experimental Psychology, 1998, 51A, (4), pp. 819-852
[16] Baddeley AD, Logie RH: 'Working memory: The multi-component model'.
in: Miyake A, Shah P, (Ed.): 'Models of working memory' (Cambridge
University Press, 1999), pp. 28-61
[17] Vandierendonck A, De Vooght G, Van der Goten K: 'Interfering with the
central executive by means of a random interval repetition task'. Quarterly
Journal of Experimental Psychology: Human Experimental Psychology,
1998, 51A, (1), pp. 197-218
[18] Pollack I, Johnson LB, Knaff PR: 'Running memory span'. Journal of
Experimental Psychology, 1959, 57, (3), pp. 137-146
[19] Zijlstra FRH, Van Doorn L: 'The construction of a scale to measure
perceived effort'. Delft, The Netherlands: Department of Philosophy and
Social Sciences, Delft University of Technology; 1985.
11
Drivers’ distraction due to ITS use
NEURAL BASIS FOR SOME COGNITIVE RISK OF
USING MOBILE PHONES DURING DRIVING
Takashi Hamada (AIST, Japan)
hamada-takashi@aist.go.jp
ABSTRACT: Two studies have been conducted on some cognitive risks of
using mobile phones during driving. Our previous study showed that voices
received by a mobile phone in a car are often interrupted and replaced by
silence due to transmission errors. We found with magnetoencephalography
(MEG) that such interruptions of voices activate the right parietal cortex, which
suggests that auditory attention can not be fully allocated for driving at moments
of the interruptions. Secondly, base on evidence that the dorsal part of the
visual system treats ‘where’ aspects of information and the ventral ‘what’
aspects, we assumed that driving in a situation is predominated by either of the
subsystems. Besides, hearing through a mobile phone was assumed to induce
visual imageries which are predominated in a situation by either of the
subsystems. Subjects thereby concurrently carried out simplified visual and
auditory tasks, each with either ‘where’ or ‘what’ aspects. Reaction times were
found to be longer when aspects in the auditory task were the same as those in
the visual task than when different.
1 General introduction
Uses of mobile phones during driving are often risky [1-8]. As a matter of fact, if
a hand is used for manipulating a mobile phone, it can not be used for driving; If
the eyes gaze at the phone, they can not be directed to the scene in front of the
car; If the cognitive resources are used for communicating with a mobile phone,
they can not be fully allocated for driving. Thus, the risks of using mobile
phones during driving are composed of manual, ocular and cognitive factors.
Although the cognitive factors are obviously based on neural activities in the
brain, they are only rarely studied from the point of neuroscience.
We studied the neural basis for some of the cognitive factors. Firstly, based on
the evidence that voices through mobile phones are often interrupted and
replaced by silence in a moving car [9], we asked how the interrupted voices
would burden the listener’s brain [10]. Secondly, a dichotomy of the visual
system that the dorsal subsystem in the parietal cortex treats spatial or ‘where’
aspect of visual information and the ventral in the inferior temporal cortex for
‘what’ aspect such as colors and shapes [11, 12] was applied for elucidating
one of the cognitive factors. Our assumptions were that driving in a situation
would be predominated by either ‘where’ or ‘what’ aspects and that hearing
through a mobile phone would induce visual imageries which would also be
predominated in a situation by either of the two aspects.
13
Human Centred Design for Intelligent Transport Systems
2 Brain activities to interrupted voices
2.1
Introduction
As we experience ourselves, voices through mobile phones are sometimes
difficult to be heard. We previously found that this is because the voices are
often contaminated by 3 types of noises: delay of transmission, distraction of
their spectral structure and silent interruptions [9]. These noises are
characterized in common by their sudden starts and sudden ends, and by their
durations of several hundreds ms, which presumably result from occasional
drops or distortion of packets (compressed segments) of voices while
processed digitally [14-16]. We previously found that long voices (vowel)
transmitted to a mobile phone in a car were interrupted more frequently, if the
car was moving (6.4 times per minutes on average) than at rest (4.8 times per
minutes on average); The average of their durations was 424 ms when the car
was moving [9]. We asked how these interruptions of voices affect the human
brain. We thereby used magnetoencephalography (MEG), which allows to
record the magnetic counterpart of the electrical activity in the brain and has
spatial and temporal resolutions finer than those in other measures such as
behavioural outputs and heart rates.
2.2
Methods
Eleven right-handed subjects listened to voices or pure tones, some parts of
which were interrupted and replaced by silence for either 200, 500 or 1000 ms
at 10 times a minute on average (Fig.1a). Their brain activities were measured
with MEG, whose signals were averaged with respect to starts of the
interruptions and then analyzed with multiple-dipole model [17].
14
Drivers’ distraction due to ITS use
Fig.1.a: Interruptions of voices and times of occurrence of two of the three
dipoles, one (T) in the temporal cortex and the other (P) in the right parietal
cortex, b: The position and direction of the parietal dipole (P). The dipole in the
right temporal cortex (T) is also shown in the sagittal section of the brain
(upper right).
2.3
Results and discussion
The interruptions of voices elicited neural activities after starts and ends of, but
not during, the interruptions. The activities after the starts could be explained by
three dipoles in the brain: One in each of the left and right temporal (auditory)
cortices (T) and one in the parietal cortex (P) of the right hemisphere (Fig.1b).
Ends of the interruptions elicited T’s in the temporal cortices, but not P in the
right parietal cortex. Interruptions of pure tones also elicited T’s, but not P.
Thus, T’s could be due to offsets or onsets of acoustic energy, while P due to
offsets of semantic flow. Previous studies [18, 19] related the right parietal
cortex to auditory attention. Since the dipole P was activated after starts of the
interruptions, but neither during nor after the interruptions, we hypothesize that
auditory attention is solicited by sudden stops of semantic flow, rather than by
waiting or by attempts to fill the message etc.
Although vision is essential for driving, audition is also required for driving. As
an example, the in-vehicle information systems often utilize auditory cues for
navigation and alerting [20, 21]. Audition is also constantly required for the
drivers to estimate the vehicle speed [22] and to detect events back of the car. If
15
Human Centred Design for Intelligent Transport Systems
a driver listens to voices through a mobile phone during driving, a part of his
resource for auditory attention should be used by the listening to the phone. The
remaining part of the resource should be further reduced at moments when
voices through a mobile phone are interrupted, which could become a risk for
driving.
3 Concurrent visual and auditory tasks with ‘where’ and
‘what’ aspects
3.1
Introduction
Although driving tasks are complex which require both of the dorsal and the
ventral visual subsystems (Fig.2), there should be a predominance from one to
the other according to a particular situation. For example, the dorsal subsystem
would be predominantly used for having the car within the lane or keeping the
distance between the cars, i.e. for controlling the car based on locations of
visual objects. In other situations, the ventral subsystem would be predominated
for identifying obstacles or colors of traffic signals. Besides, hearing through a
mobile phone could sometimes induce visual imagery which would again
predominantly use either of the visual subsystems depending on situations. For
instance, if a partner of the phone explains the place of Institut Lumière by
saying "For reaching there, you start from the station Part-Dieu, go straight two
blocks, turn to the right and then to the left...", the listener should then mentally
image the visual map of the route; The dorsal visual subsystem would then be
activated [23]. In another situation, if the partner of the phone says “Her dress
yesterday was beautiful, wasn’t it?”, the listener should then visually image its
colors; The ventral visual subsystem would then be activated [24]. Thus, the
concurrent driving and hearing through a mobile phone in a situation would be
schematically reduced to neural activations in one of the four combinations, i.e.
two (dorsal or ventral) through vision multiplied by two (dorsal or ventral)
through audition. We asked which of the combinations would be more
dangerous than the others.
Fig.2. Dorsal (D) and ventral (V) visual subsystem. A: auditory cortex, PV:
primary visual cortex.
16
Drivers’ distraction due to ITS use
3.2
Methods
Subjects (n=7) concurrently carried out visual and auditory tasks (Fig.3), while
their MEG were recorded. A spot light was presented either as red or green at
the center of the visual field in the visual color task (vC), whereas at the left or
the right position in the visual location task (vL). They differentially pushed one
of the two buttons according to the aspects of the visual stimuli with a finger of
his right hand (the left button to green or left spots and the right button to red or
right spots) . Reaction times (RTs) of pushing the buttons were recorded. In a
trial of the auditory color task (aC), they heard the name of a colored object,
such as "apple" and "cabbage", and had to push a button with his left hand if
the object was red, but not if green. In the first trial of the auditory location task
(aL), they visually imaged a pointer at the center of a 3x3 imaged matrix. In
each of the trials, they heard a direction word, "left", "right", "up" or "down", and
accordingly moved the pointer for one step in the matrix. If the pointer exceeded
the matrix, they pushed the button with a finger of his left hand and the trials
were reset. Otherwise they only memorized the location of the pointer in the
matrix for the next trial.
Fig.3. Time sequences of the audiovisual tasks
In an experimental session, one of the auditory tasks (aC or aL) and one of the
visual tasks (vC or vL) were combined. Onsets of the former with respect to the
latter (tAV) were randomly varied between -500 and -250 ms from trial to trial in a
session. The subjects carried out the visual tasks with priority over the auditory
tasks. Two further sessions only with visual stimuli (vL or vC) were carried out
as the controls. More than 250 trials were repeated in a session. Errors in
pushing the buttons occurred in trials less than 5 %; RTs in such trials were
excluded from analysis. RTs at tAV of -500 and -250 in a session of a subject
were combined and finally grand-averaged across the subjects, which were
statistically compared among the different sessions.
3.3
Results and discussion
Since MEG signals were too complicated to be analysed, only behavioral
results are described. Fig.4a shows that RTs in the audiovisual sessions were
always longer than that in the control, which suggests that concurrent hearing is
17
Human Centred Design for Intelligent Transport Systems
dangerous for driving. RTs in aLvL and RTs in aCvL were subtracted by RT in
vL in each of the subject. The results represent to what extent RTs in vL were
increased by the concurrent auditory tasks (aL or aC) in the subject. These
increased amounts were named RT*(aLvL) and RT*(aCvL) of the subject,
respectively. Similarly, RTs in aLvC and RTs in aCvC were subtracted by RT in
vC, and each of the results was named RT*(aLvC) and RT*(aCvC) of the
subject. Grand-averages of these increased amounts are represented by two
pairs of two bars (Fig.4b). The figure shows that RT*(aLvL) are significantly
longer than RT*(aLvC) (312 ms vs 192 ms, respectively; p<0.05), and
RT*(aCvC) are also significantly longer than RT*(aCvL) (152 ms vs 64 ms,
respectively; p<0.01).
Fig.4.a: Grand-averaged RTs in different sessions. Error bars indicate SD. Two
bars connected by a horizontal line have statistically different heights. **:
p<0.01, *: p<0.05. b: Grad-averaged RTs in the audiovisual sessions subtracted
by RT in the corresponding control session. Others are the same as in a.
In summary, RTs in the visual tasks were more elongated when the concurrent
auditory task had the same aspect as the visual task than when different. The
dichotomy of the human visual system can account for these results (Fig.5). As
an example, since both the tasks in aLvL are supposed to use the dorsal
subsystem, the subsystem could not be fully allocated for the visual task vL. In
contrast, in the session of aLvC, since the auditory task aL is supposed to use
the dorsal subsystem and the visual task vC the ventral, the ventral subsystem
could be fully allocated for the visual task vC. These considerations could
explain why RTs were more increased by addition of the auditory task in aLvL
than in aLvC. Similar arguments apply for the difference of RTs between the
sessions of aCvC and aCvL. The results implicate that hearing through a mobile
phone during driving would be more dangerous when aspects of information
with which hearing and driving should deal are the same than when different.
18
Drivers’ distraction due to ITS use
Fig.5. Neural explanations of the results
4 General discussion
Two studies tried to elucidate cognitive risks of using mobile phones during
driving from the point of neuroscience. The first study used MEG to assess the
impact of phone communication interruptions to the brain. The results showed
that the right parietal cortex was activated after starts of the interruptions, which
suggests that the neural resource for auditory attention could then be less
allocated for driving. Behavioral experiments in realistic situations are required
for validating this hypothesis. The second study was based on the dichotomy of
the human visual system: dorsal subsystem for “where” aspects of information
and ventral for “what” aspects, as a new type of the multiple resource models
[25]. The experiments used reaction times to asses the possible interference on
visual tasks by visual imagery induced by the auditory communication. The
results implicate that concurrent driving and hearing through a mobile phone
could be more dangerous if aspects of information with which driving and visual
imagery induced by hearing should deal are of the same type, i.e. both of color
or both of location, than when different.
In the first experiment, brain activities to the interruptions were successfully
described by dipoles. However, MEG signals in the second study were too
complicated to be analyzed, presumably because the experimental setup was
not simple enough. Neural studies have advantages that they have finer spatial
and temporal resolutions than behavioral or psychophysiological studies, but
have disadvantages that they generally require simplified, not realistic, setups.
Thus, these different types of studies have to be combined for fully analyzing
cognitive risks of using mobile phones during driving.
19
Human Centred Design for Intelligent Transport Systems
5 References
[1]
[2]
[3]
[4]
[5]
[6]
[7]
[8]
[9]
[10]
[11]
[12]
[13]
[14]
[15]
[16]
[17]
Alm, H. and Nilsson, L. :The effect of a mobile telephone task on driver
behaviour in a car following situation, Accid. Analy. and Prev., 1995, 27,
pp. 707-715
Brookhuis, K. A., de Varies, G. and de Waard, D.: The effect of mobile
telephoning on driving performance, Accid. Analy. and Prev., 1991, 23, pp.
309-316
Brown, I.D., Tickner, A.H. and Simmonds, D.C.V.: Interference between
concurrent tasks of driving and telephoning, J. Appl. Psych., 1969, 53, pp.
419-424
Consiglio, W., Driscoll, P., Witte, M. and Berg, P.: Effect of mobile
telephone conversations and other potential interference on reaction time
in a braking response, Accid. Analy. and Prev., 2003, 35, pp. 495-500
Laberge-Nadeau, C., Maag, U., Bellavance, F., Lapierre, S.D.,
Desjaardins, D., Messier S. et al.,: Wireless telephones and the risk of
road crashes, Accid. Analy. and Prev., 2003, 35, pp. 649-660
Törnros, J. and Bolling, A.: Mobile phone use – effects of conversation on
mental workload and driving speed in rural and urban environments,
Trans. Res. F, 2006, 9, pp. 298-306
Strayer, D.L., Drews, F.A., Crouch, D.J. and Johnston, W.A.: Cell phone
and driving, in Walker, W. R. and Herrmann, D. (Eds.): ‘Cognitive
Technology: Transforming Thought and Society’ (McFarland & Company
Inc., Jefferson, NC. 2006)
Horrey, W.J. and Wickens, C.D.: Examining the impact of cell phone
conversations on driving using meta-analytic techniques, Hum. Fact.,
2006, 48-1, pp. 196-205
Kawano, K., Iwaki, S., Azuma, Y., Moriwaki, T. and Hamada, T.: Degraded
voices through mobile phones and their neural effects: A possible risk of
using mobile phones during driving, Trans. Res. F, 2005, 8, pp. 331-340
Hamada, T., Iwaki, S. and Kawano, T.: Speech offsets activate the right
parietal cortex, Hearing Research, 2004, 195, pp. 75-78
Jacob, P. and Jeannerod, M. : ‘Ways of seeing’ (Oxford University Press,
2003)
Pisella, L., Mohammad, A., and Rossetti, Y.: The timing of color and
location processing in the motor context, Exp. Brain Res., 1998, 121, pp.
270-276
Hamada, T., Trans. Res. F, 2008, 11, pp. 75-82.
Enderes, T., Khoo, S.C., Somerville, C.A. and Samaras, K.: Impact of
statistical multiplexing on voice quality in mobile networks, Mobile networks
and applications, 2002, 7, pp. 153-161
Janssen, J., De Vleeschauwer, D, Buchli, M. and Petit, G. H.: Assessing
voice quality in packet-based telephony, IEEE Internet Comput., 6, pp. 4856
Robinson, H. M.: Voice quality, Packet, 2002, 4, pp. 15-17
Hämäläinen, M., Hari, R., Ilmoniemi, R. J., Knuutila, J., and Lounasmaa,
O. V. : Magnetoencephalography - theory, instrumentation, and
20
Drivers’ distraction due to ITS use
[18]
[19]
[20]
[21]
[22]
[23]
[24]
[25]
applications to noninvasive studies of the working human brain, Rev.
Modern Phys., 1993, 65, pp. 413-497
Kasai, K., Nakagome, K., Itoh, K., Koshida, I., Hata, A., Iwanami, A.,
Fukuda, M., Hiramatsu, K. I. and Kato, N.: Multiple generators in the
auditory automatic discrimination process in humans, NeuroReport, 1999,
10, pp. 2267-2271
Levänen, S., Ahonen, A., Hari, R., McEvoy, L. and Sams, M.: Deviant
auditory stimuli activate human left and right auditory cortex differently,
Cereb. Cortex, 1996, 6, pp. 288-296
Liu, Y.-C.: Comparative study of the effects of auditory, visual and
multimodality displays on drivers’ performance in advanced traveler
information systems, Ergonomics, 2001, 44, pp. 425-442
Wiese, E. E. and Lee, J. D.: Auditory alerts for in-vehicle information
systems: The effects of temporal conflict and sound parameters on driver
attitudes and performance, Ergonomics, 2004, 47, pp. 965-986
Evans, L.: Speed estimation from a moving automobile, Ergonimics, 1970,
13, pp. 219-230
Mellet, E., Tzourio, N., Crivello, F., Joliot, M., Denis, M. and Mazoyer, B.:
Functional anatomy of spatial mental imagery generated from verbal
instructions, J. Neurosci., 1996, 16-20, pp. 6504-6512
Howard, R.J., ffytche, D.H., Barnes, J., McKeefry, D., Ha Y., Woodruff,
P.W., Bullmore, E.T., Simmons, A., Williams, S.C., David, A.S. and
Brammer, M.: The functional anatomy of imaging and perceiving colours,
NeuroReport, 1998, 9, pp. 1019-1023
Wickens, C. D.: Multiple resources and performance prediction,
Theoretical Issues in Ergonomics, 2002, 3, pp. 159-177.
21
Drivers’ distraction due to ITS use
EFFECTS OF VISUAL SEARCH TASK COMPLEXITY
ON LANE CHANGE TASK PERFORMANCE
Ellen S. Wilschut (IfADo, Germany),
Gerhard Rinkenauer (IfADo, Germany),
Karel A. Brookhuis (Groningen University, The Netherlands),
Michael Falkenstein (IfADo, Germany)
wilschut@gmx.de
ABSTRACT: Driver assistance systems are increasingly implemented in motor
vehicles. However, it is unclear whether the secondary tasks introduced by
these systems affect driving performance and whether they cause safety risks
to the driver and road traffic. In this study the effect of secondary task
complexity on driving performance is manipulated using different complexity
manipulations of a visual search task. The lane-change task [19] was used as
the primary task to simulate driving. Results showed that participants (n=12)
were unable to maintain their baseline driving performance when the secondary
task had to be performed. Moreover, they showed further dual task decrements
with increasing visual search complexity. The results show adverse effects on
simulated driving depending on the complexity of an additional task.
1 Introduction
The most frequently implemented in-vehicle information systems (IVIS)
nowadays are the In-Vehicle Routing and Navigation Systems (IRANS) that
provide drivers information about the route from one destination to another [4].
Research has shown that under some circumstances dual task decrement can
affect driving performance [5, 6, 14, 15, 23]. However, people are engaged in
concurrent tasks on a daily basis, like drinking coffee while watching the news.
Most of these tasks are highly practised and are therefore highly automated
[22]. People are fairly good at performing multiple tasks at the same time
provided at least one of them is (highly) automated, and in case the execution
of a task threatens to fail, people have a large range of coping mechanisms
available. There are at least three ways in which drivers can adapt their
behaviour to cope with higher task demand [3, 13]. These are: investment of
more effort, changing working strategy and neglecting subsidiary information. In
a study of Dingus et al. [9] IRANS with the highest visual demand were
associated with the lowest driving speed. Thus drivers adapted their working
strategies and made the driving task less demanding by lowering their speed
[see also 12]. It is expected that driver will adjust their strategies to deal with
this additional task demand and reach what can be described as homeostasis
or an optimum level of accepted risk or task difficulty [11, 12, 27]. For instance,
Pohlmann and Traenkle [21] also found speed reductions and a deterioration in
lateral control with high visual demand of IRANS. They found this effect
particularly near intersections. Drivers reduce speed to allow time to drive safely
and were highly motivated to check the IRAN system, even in difficult traffic
situations. Unfortunately, it is especially near intersections which are complex
traffic situations that the need for route information is high [8]. So this could be a
23
Human Centred Design for Intelligent Transport Systems
situation where task demands are higher than normal and possibly cause dual
task decrement. As long as the driving task is self-paced and compensating
strategies can be executed the interference of secondary tasks will be limited.
However, driving can also be paced by the environment. For instance driving 30
km/h on the highway can be done theoretically, but is certainly more hazardous
than missing the highway exit. In that case compensating by considerably
reducing speed would hardly be possible. So there are situations in which
compensating strategies are not sufficient or can not be executed, and when
this is the case and task demand is high, driving performance is likely to suffer.
1.1
Visual search
Using a digital map requires searching and detecting relevant information. If the
system is well-designed this search for relevant information is efficient, requiring
the driver to take the eyes off the road for a minimum amount of time. Complex
colouring, multiple signs and long text messages in displays make the search
for information demanding and inefficient, and require multiple glances. Factors
that influence the search time of a display have been widely studied using all
kinds of visual search paradigms. Visual search is easier when the target can
be defined by one feature such as colour, e.g. the target colour (blue) is
different from its distractor (red). Within the Feature Integration Theory [25] this
type of feature search is called pop-out. When a target is defined by a
conjunction of two or more features, for instance the target is a blue letter ’A’
between distractors consisting of blue ‘T’s and red ‘A’s, search for a target is
slower. The time needed to search for a conjunction target becomes slower as
a function of the set size (the number of elements in a search display) while
search for a pop-out target will be relatively independent of set size. Theories
on attention try to explain this difference in reaction times [18] between
searching for conjunction versus pop-out targets. The fundamental nature of
this attentive selection process is still under debate to date [10, 17, 25], but
clear is that there is not such a clear dichotomy in visual search as desribed by
Treisman, therefore Wolfe [28] suggested to use the term “efficiency” to
describe the continuum of search complexity. The efficiency of visual search
depends for instance on the number of distractors and on the number of
features in the display (e.g. form and colour). A typical finding is that more
distractors and more equality between distractors and targets makes the visual
search less efficient, slower and less accurate [24, 29]. Interesting research has
been conducted in the HASTE project [20] in which a visual search task is used
as a surrogate IVIS. The visual search task was performed both in isolation and
concurrently with a laboratory driving task, in a driving simulator and while
actually driving on the road. The visual surrogate IVIS task of HASTE consisted
of a choice-reaction task with three difficulty levels as shown in figure 1. Each
display contained a mixture of pop-out and conjunction displays i.e. of the
different classes of search requirements. Results showed that the difficulty of
the visual task had a pronounced effect on steering (a higher steering reversal
rate) and lateral behaviour (higher standard deviation of the lateral position).
Also the increased secondary taks load let to speed reduction and increase in
headway.
24
Drivers’ distraction due to ITS use
Difficulty level 1
Difficulty level 2
Difficulty level 3
Fig.1. Example displays of the surrogate IVIS displays. Within each difficulty
level there are in fact different types of displays. For instance in difficulty level 2
one fourth of the displays are pop-out displays (the upward pointing target arrow
‘pops out’ in left display). And the other part of displays require conjunction
search.
1.2
Aim of the study
The aim of the present study is to extend the previous research by focusing on
the effect of search difficulty (pop-out versus conjunction) and the effect of set
size. Another factor that differentiates the present study from the HASTE
experiments is that driving speed was fixed at 60 km/hour. Thus participants
were unable to compensate for the task demand by decreasing their speed. In
the present study the effect of secondary, visual search complexity on driving
performance is evaluated, using a visual search task as a surrogate IVIS and
the lane-change task as the primary task to simulate driving [19]. Both tasks
were performed separately to acquire the baseline values of each participant,
and are then compared to the performance of the two tasks in a dual task
situation. In this dual task situation the lane-change task was defined as the
primary task which is instructed to be given the highest priority. It was expected
that in the dual task situation the performance of the visual search task would
decrease and the secondary task may have a possible negative effect on the
simulated driving task if the participants are unable to fully prioritise the LCTtask or compensate for the dual task demand. We expect this effect to be larger
with increasing set size and with conjunction search displays.
25
Human Centred Design for Intelligent Transport Systems
2 Method
2.1
Participants
Twelve female participants aged between 20 and 22 years (M=20.3 (±0.6))
participated in a experimental session in exchange for a partial course credit. 1
All participants had their driver's licence for at least 2 years and had normal or
corrected to normal vision. The experimental procedure consisted of three
different parts. The first part was a simulated driving task called the lane change
task [19]. The second part was a visual search task. The sequence of these two
parts was counterbalanced across participants. The final part of the experiment
combined the simulated driving and the visual search task. Each part started
with a training block. In total participants received about half an hour of training.
The training of the visual search task continued until the participants reached a
minimum of 80% correct trials. Each part was followed by a short break.
2.2
Primary simulated driving task
The simulated track consisted of a straight three-lane road. With the gas pedal
pressed maximally the participant drove a constant 60 km/h. This resulted in a
total driving time per track of about 3 minutes. There were 18 signs along each
track indicating the lane the participant had to change to as soon as the sign
was identified. Signs were present with a mean distance of 150 m (min. 140 and
max. 188 metres, exponentially distributed). Each of the 6 possible lane
changes occurred three times during one track. The performance on six
different tracks was measured of which the first two were considered training.
Data were continually sampled with a scan rate of 130 Hz, namely the lateral
and longitudinal position, speed (as a control) and steering angle. As a measure
of driving performance the deviation between a normative model and the
participants’ actual course on the track was calculated [19].
2.3
Secondary visual search task
In the visual search task each experimental trial began with the presentation of
a fixation cross, which remained on screen for 1000 ms. The fixation cross was
followed by the presentation of a visual search display for 3000 ms. The display
types consisted of a complexity manipulation pop-out/conjunction and a set size
manipulation (4 versus 9 items). The four types of display were blocked and
randomly presented to the participants. Participants were presented with 8
blocks, each consisting of 40 trials. The target stimulus was either an upward
green arrow or a right-pointing red arrow, only one target was present per trial
with a 50% change. The target appeared on a random location. The non-targets
were all other combinations of green or red arrows with orientation to the left, to
1
Since this experiment was part of an university course, therefore participants
were not divided equally across gender. Adam [1]showed that female spatial
choice-RTs were about 34 ms slower than those of male participants and
different strategies were adopted per gender group, effects like this should be
considered when determing the external validity of the study.
26
Drivers’ distraction due to ITS use
the right, up or down. The participant had to indicate by a button press whether
the target was present or not. They were instructed to react as quickly and
accurately as possible. In the pop-out displays the target was distinguishable
from the non-target either by colour or orientation. Thus there were two possible
situations: distractor arrows all pointed in the vertical direction when the target
was a red right-pointing arrow or when the target was a green upward arrow, all
distractors were red. In conjunction displays the non-targets pointed in all
directions and the amount of red and green arrows was equal.
1
Fig.2. Four example displays with the two different target arrows (displays are
described from left to right): The two left displays are the pop-out search
displays; participants can discriminate target arrows from distractor arrows by
looking at direction or colour. In display 1: a red arrow (= grey) pointing to the
right or in display 2: a green (= white) upward pointing arrow. The displays on
the right side are the conjunction search displays, where both features colour
and direction have to be determined to find the target. The same targets are
presented in display 3: a red arrow pointing to the right and display 4: a green
upward pointing arrow.
2.4
Dual task
In the dual task the instructions and conditions for driving as well as for the
visual search task were kept equal and the participant was instructed to give
first priority to the driving task. The constant, required driving speed of 60 km/h
made it impossible to compensate for secondary task difficulty by reducing
speed. In each of the eight blocks one of the four visual search conditions was
presented. Each condition was thus presented twice and the order was random.
2.5
Equipment and analysis
For tracking a simple Logitech gaming steering wheel was used with gas and
brake pedals. The visual search task was displayed on a small LCD screen at a
distance of 1.45 m with a visual angle of 18°. This LCD screen was positioned
in front of the main screen (distance 1.96 cm, visual angle 38° x 29°) without
blocking the sight on the road. For stimulus presentation E-prime was used
(Psychology Software Tools, Inc., Pittsburgh, USA) the small set size was
visible within 1°, the large set size within 1.6°. Two of the original response
buttons behind the steering wheel were adapted to make accurate RT
acquisition possible using the printer port. Data were subjected to ANOVA
repeated measurement analyses of SPSS 13.0, except for training blocks. In
case of sphericity violation the Greenhouse-Geiser modification was used.
Within-subjects factors were complexity (pop-out/conjunction), set size (4/9
items) and single versus dual task performance. For the behavioural results of
27
Human Centred Design for Intelligent Transport Systems
the visual search task target and non-target reactions are combined in the
analysis.
3 Results
3.1
Visual Search
The analysis of the RT showed that participants took on average 86 ms longer
to react in the dual task situation (F(1,11)=4.3, p=.063; see table 1). RTs were
about 1000 ms slower for conjunction displays where participants had to search
for a target looking at its two features colour and rotation (F(1,11)=425.8,
p<.001; see figure 3a). Participants were reacting on average 260 ms slower for
the large (nine elements) displays then for the small (four elements) displays
(F(1,11)=194.1, p<.001). Also the expected interaction between complexity and
set size of the visual search task reached significance (F(1,11)=110.0, p<.001)
showing a larger increase of RTs with set size when the displays were
conjunction displays.
2500
50
A
40
Error (%)
RT (ms)
2000
1500
1000
B
30
20
10
500
0
0
4
4
9
9
set size
set size
Single task pop-out
Dual task pop-out
Single task conjunction
Dual task conjunction
Pop-out
Conjunction
Fig.3. A) Reaction times of the visual search task in the dual task condition
(averaged over target and non-target displays) for the conditions set size and
efficiency. B) Errors in percentage of the visual search task in single and dual
task condition (averaged over number of false alarms and misses). The
increase of the percentage of errors with dual task can be attributed to the
increase in the number of misses.
Participants hardly made any errors with pop-out displays, i.e. 0.2% compared
to 16.5% with conjunction displays (F(1,11)=96.6, p<.001). Also with larger set
size the amount of errors increased (F(1,11)=19.2, p<.001). And the interaction
between set size and the complexity of search was significant showing the
same direction as RTs, i.e. the amount of errors increased more with set size
when the displays were conjunction (F(1,11)=13.1, p<.004). Participants made
significantly more errors in the dual task condition; 8.5% in single task
performance versus 10.6% when performing the visual search task together
28
Drivers’ distraction due to ITS use
with the lane change task (F(1,11)=6.2, p<.030). Within the dual task the
number of errors increased for display complexity (F(1,11)=12.2, p<.005), set
size (F(1,11)=16.0, p<.002) and also the interaction between complexity and set
size was significant (F(1,11)=24.8, p<.001). This interaction showed that when
the visual task is performed as a dual task the number of errors increases and
the errors increase more drastically for conjunction displays with a large set
size. When looking at the type of error (false alarms and misses see table 1) in
the dual task situation, there is no significant difference in the false alarms, so
the effect can be attributed to an increase in the number of misses in the dual
task condition.
3.2
Primary task
The mean deviation from the ideal lane keeping (the normative model) showed
an effect of dual task (F(4,44)=11.9, p<.001). It showed that even in the most
simple dual task condition drivers were unable to perform the lane change task
at baseline level (F(1,11)=7.5, p<.019). The complexity of the search
(F(1,11)=12.5, p<.005) and display size (F(1,11)=5.7, p<.036) reached
significance as well as the interaction between the two conditions (F(1,11)=9.0,
p<.012). Effects showed an increase of the deviation of the ideal lane keeping
with increased set size and conjunction search (see figure 4). The variance of
steering wheel angle is reduced as an effect of dual task but this measure
showed no significant effects of the visual search manipulations. Participants
were able to keep a constant speed of 60 km/h during the whole experiment.
Table 1. Performance measures of the
visual search task
Mean deviation [m]
1.8
Pop-out
Conjunction
Baseline
1.6
1.4
RT (ms)
1.2
4
9
set size
Fig.4. Mean deviation from the
normative model for single task
baseline driving and driving with
different visual search conditions.
Error (%)
Misses
False
alarms
Dual
Task
DualSingle
difference
p
value
1359
(39.3)
10.6 (1.1)
13.4 (2.7)
20.4 (3.4)
85.9
(37.0)
2.1 (0.8)
9.1 (2.1)
-1.3 (2.6)
.063
.030
<.001
n.s.
Note: Displayed are the mean and standard
error. Performance measures on dual task
minus single task are given and the paired ttest of this dual vs. single task difference.
4 Discussion
People are fairly good at performing multiple task at the same time when one of
the task is automated or highly trained or when coping mechanisms can be
applied when task performance is threatened One of the common adaptations
seen while driving is to reduce the speed to allow more time for decision making
[9, 12, 21, 27]. Because the speed in the experiment was fixed there was no
29
Human Centred Design for Intelligent Transport Systems
possiblity to compensate for high task demand by reducing the speed. The
results of this experiment showed that drivers were unable to continue
performing the driving task adequately at baseline level in the dual task
condition. Apparently they could not or at least did not sufficiently prioritise the
lane change task. Like in the HASTE studies [20] where higher complexity of
the visual search task resulted in a large deviation of the lateral positioning on
the road, the performance on the lane change task decreased, as was shown in
the mean deviation from the normative model. The driving decrement was found
even for the most simple search displays containing only four elements with a
pop-out target based on colour or orientation. The visual search task
performance showed was in accordance with the literature [7, 24, 25, 29], i.e.
almost no increase in RT or errors for pop-out displays when set size increases,
but a large increase of RT and errors with setsize for conjunction displays. The
visual search task also affected the mean deviation of the driving task.
Participants were unable to perform the secondary, visual search task at
baseline level while driving making substantially more errors. Although there
was only a trend (p<.063) showing that reaction time increased in the dual task
condition, the number of errors increased most strongly with large conjunction
displays. The amount of errors could be attributed to the increase of the number
of misses. A likely explanation for the number of errors is that the participants
did not have enough time to inspect the visual display for the target.
Although it is not directly verifiable it could be the case that the larger mean
deviation from the ideal driving track resulted from participants spending a
larger amount of their time looking at the visual search display or taking a long
time switching between the two tasks. To minimize the physical switching
distances between two objects of visual attention, head-up displays (HUDs)
have been introduced into the modern automobile. The benefit of this
technology is that it decreases eyes-off-the-road and accommodation time [16],
although there are also some concerns about the cluttering of information [26].
A follow-up of the present experiment is currently being performed to investigate
the use of a HUD to verify if the performance decrease in the driving task can
be attributed to the time that the participants spend looking on the display as
opposed to divided attention. Another follow up on this study will focus on the
elderly driver because they could be especially vulnerable to dual task
interference, having motoric perceptual and cognitive functions decline due to
normal ageing which affects driving [2, 20].
When using a laboratory driving task to measure the driving performance it
should be considered that this environment is artificial and can give a false
impression of driving performance. Further results can not be assumed to
transfer exactly to driving on the road. It can be concluded that when
participants can not adapt their speed, dual task performance decreases when
displays require conjunction search and a large number of elements are
displayed. Displays with conjunction features forcing inefficient search should
be avoided, because reaction time increases and in the dual task situation the
amount of errors increases drastically, and at least simulated driving
performance is worsened.
30
Drivers’ distraction due to ITS use
5 Acknowledgments
This study was funded by a grant for early stage researchers in the HUMANcentred design for Information Society Technologies EU-project (HUMANIST;
reference number 507420). We would like to thank two anonymous reviewers
for their useful comments.
6 References
[1]
[2]
[3]
[4]
[5]
[6]
[7]
[8]
[9]
[10]
[11]
[12]
[13]
Adam, J., Paas, F., Buekers, M., Wuyts, I., Spijkers, W., Wallmeyer, P.
‘Gender differences in choice reaction time: evidence for differential
strategies’. Ergonomics. 1999;42(2), pp. 327-35.
Anstey, K.J., Wood, J., Lord, S., Walker, J.G. ‘Cognitive, sensory and
physical factors enabling driving safety in older adults’. Clin Psychol Rev.
2005 Jan;25,(1), pp. 45-65.
Bainbridge, L.’Problems in the assessment of mental load’. Le Travail
Human. 1974;37, pp. 279-302.
Barfield, W., Dingus, T.A. ‘Human Factors in Intelligent Transportation
Systems.’ Mahwah, New Jersey: Lawrence Erlbaum Associates; 1998.
Brookhuis, K.A., deVries, G., deWaard, D. ‘The effects of mobile
telephoning on driving performance.’ Accid Anal Prev. 1991;23,(4) , pp.
309-16.
Brouwer, W.H., Waterink, W., Van Wolffelaar, P.C., Rothengatter, T.
‘Divided attention in experienced young and older drivers: lane tracking
and visual analysis in a dynamic driving simulator.’ Hum Factors. 1991
Oct;33,(5), pp. 573-82.
Chelazzi, L. ‘Serial attention mechanisms in visual search: a critical look at
the evidence.’ Psychol Res. 1999;62,(2-3), pp. 195-219.
Cnossen, F., Meijman, T., Rothengatter, T. ‘Adaptive strategy changes as
a function of task demands: a study of car drivers.’ Ergonomics. 2004 Feb
5;47,(2), pp. 218-36.
Dingus, T., Hulse, M., Mollenhauer, M.A., Fleishman, R.N., McGehee,
D.V., Manakkal, N. ‘Effects of age, system experience, and navigation
technique on driving with an advanced traveler information system.’
Human Factors. 1997;39,(2), pp. 177-99.
Duncan, J. ‘The locus of interference in the perception of simultaneous
stimuli.’ Psychological Review. 1980;87, pp. 272-300.
Fastenmeier, W., Gstalter, H. ‘Driving task analysis as a tool in traffic
safety research and practice.’ Safety Science. 2006;In Press, Corrected
Proof.
Fuller, R. ‘Towards a general theory of driver behaviour.’ Accident Analysis
& Prevention. 2005;37,(3), pp. 461-72.
Hockey, G.R.J. ‘Compensatory control in the regulation of human
performance under stress and high workload: a cognitive-energetical
approach.’ Biological Psychology. 1997;45, pp. 73 – 93.
31
Human Centred Design for Intelligent Transport Systems
[14] Horrey, W.J., Wickens, C.D. ‘Examining the impact of cell phone
conversations on driving using meta-analytic techniques.’ Hum Factors.
2006 Spring;48,(1), pp. 196-205.
[15] Janssen, W.H., Kaptein, N., Claessens, M. ‘Behavior and safety when
driving with in-vehicle devices that provide real-time traffic information.’
Toronto; 1999.
[16] Liu, Y., Wen, M. ‘Comparison of head-up display (HUD) vs. head-down
display (HDD): driving performance of commercial vehicle operators in
Taiwan.’ International Journal of Human-Computer Studies. 2004;61,(5),
pp. 679-97.
[17] Madden, D.J., Spaniol, J., Bucur, B., Whiting, W.L. ‘Age-related increase in
top-down activation of visual features.’ Q J Exp Psychol (Colchester). 2007
May;60,(5), pp. 644-51.
[18] Maquestiaux, F., Hartley, A.A., Bertsch, J. ‘Can practice overcome agerelated differences in the psychological refractory period effect?’ Psychol
Aging. 2004 Dec;19,(4), pp. 649-67.
[19] Mattes, S. ‘The Lane-Change_Task as a tool for Driver Distraction
Evaluation.’ Book: Quality of work and Products in Enterprises of the
Future. 2003, pp. 57-60.
[20] Merat, N., Anttila, V., Luoma, J. ‘Comparing the driving perfomance of
average and older drivers: The effect of surrogate in-vehicle information
systems.’ Transportation Research Part. 2005;F 8, pp. 147-66.
[21] Pohlmann, S., Traenkle, U. ‘Orientation in road traffic. Age-related
differences using an in-vehicle navigation system and a conventional map.’
Accident Analysis & Prevention. 1994;26,(6), pp. 689-702.
[22] Rasmussen, J. ‘Skills, Rules, and Knowledge; Signals, Signs, and
Symbols, and Other Distinctions in Human Performance Models.’ IEEE
Transactions on Systems, Man, and Cybernetics. 1983;13,(3), pp. 257-66.
[23] Santos, J., Merat, N., Mouta, S., Brookhuis, K.A., de Waard, D. ‘The
interaction between driving an in-vehicle information systems: Comparison
of results from laboratory, simulator and real-world studies.’ Transportation
Research Part. 2005;F 8, pp. 135-46.
[24] Treisman, A. ‘Perceptual grouping and attention in visual search for
features and for objects.’ Journal of Experimental Psychology: Human
Perception and Performance. 1982;8,(2), pp. 194-214.
[25] Treisman, A.M., Gelade, G. ‘A feature-integration theory of attention.
Cognitive Psychology.’ 1980, pp. 12:97-136.
[26] Wickens, C.D. ‘Engineering Psychology and Human Performance.’ New
York: Harper-Collins Publishers, Inc.; 1992.
[27] Wilde, G.J.S. ‘The theory of risk homeostasis: implications for safety and
health.’ Risk Anal. 1982;2, pp. 209-25.
[28] Wolfe, J.M. ‘Visual search.’ In: Pashler HE, editor. Attention. Hove, East
Sussex: Psychology Press; 1998, pp. 13-59.
[29] Wolfe, J.M., Cave, K.R., Franzel, S.L. ‘Guided search: an alternative to the
feature integration model for visual search.’ Journal of experimental
psychology. 1989 Aug;15,(3), pp. 419-33.
32
Drivers’ distraction due to ITS use
EFFECTS OF SIMULTANEOUS MULTI-MODAL
WARNINGS AND TRAFFIC INFORMATION ON
DRIVER BEHAVIOUR
Clemens Kaufmann (FACTUM OHG, Austria),
Ralf Risser (FACTUM OHG, Austria),
Arjan Geven (CURE, Austria),
Reinhard Sefelin (CURE, Austria)
clemens.kaufmann@factum.at
ABSTRACT: It is expected that the number of in-vehicle telematic systems will
increase rapidly over the next years, leading to an increased amount of
information a driver has to deal with. In the scope of the research project
LIVES 1 , the organisations CURE and FACTUM OHG have investigated how
these systems could be used optimally to improve the safety on the road. To be
able to answer the research questions formulated by FACTUM OHG, CURE
developed a driving simulator specifically tuned to these questions. The driving
simulator made it possible to investigate the effects on the driving behaviour
using different modalities for information and the effects of simultaneously
submitted information. The results of these experiments were used to
formulated guidelines to increase safety. These guidelines focus on the optimal
use of auditory, visual and haptic information for the driver as well as on the
simultaneous communication of more than one system.
1 Starting point
It is expected that the number of in-vehicle telematic systems will increase
rapidly over the next few years. This will lead to an increased amount of
information a driver has to deal with besides the primary task of driving. The
main advantages of telematic systems are or should be the ability to increase
both the safety and the comfort of the driving task. These advantages, however,
might be compensated by additional perceptual and cognitive load, as the
driving task is extended with the task of reacting correctly on the information,
directions and warnings that these systems provide. This will especially be the
case when the driver receives different information simultaneously from these
systems.
The research institutes CURE (Center for Usability Research & Engineering)
and FACTUM OHG (Transport- and Social Analyses) investigated in the year
2006 in the Austrian study LIVES how the human machine interface (HMI) of
different telematic in-vehicle systems should be designed in order to guarantee
an optimal use of diverse systems [1]. It dealt with the integration of disjunctly
developed systems, that are not a priori matched to one another. Thus, the
driver may run the risk of being distracted by unimportant or interfering
information during possibly critical situations.
1
German title: LIVES - LenkerInnenInteraktion mit VErkehrstelematischen Systemen
33
Human Centred Design for Intelligent Transport Systems
2 Development of hypotheses
One main focus of the project was the identification of new requirements to
drivers which might appear through the implementation of telematic systems
into cars. Especially the effects on the driver and his/her behaviour while he/she
gets simultaneous information from two systems were tested. Another focus
was on the search for the optimum modality in which information should be
submitted to the drivers.
Based on a literature study several telematic in-vehicle systems and situations,
where systems submit either information (traffic news) or warnings (too little
distance to the car ahead etc.) to the driver, which suited the project objectives
best were selected. These information and warnings have different priorities
with regard to the driving task and the driver's reaction in different situations.
Three priority levels were defined:
1. Warnings which need an immediate reaction by the driver
2. Latent instruction – no immediate reaction of the driver is needed
3. General information – no immediate relevance for the driving task
One of the main questions was: "What is the best way and the best modality to
submit information or warnings in order not to distract or to overload the driver?"
Based on literature the following matrix was established, in which the priority
levels are set in contrast to the three modalities acoustic, visual and haptic.
Table 1. Classification of information into priority levels and modalities
modality
type of
information
warnings, high
priority
medium priority
general
information, low
priority
acoustic
visual
haptic
suited well, it can be
instinctively handled.
Speech is not adequate
[2]
acoustic signals are
experienced as
hindering and therefore
should not be used for
this level
not adequate, modality
is to slow to transfer the
information in critical
situation
suitable. Drivers can
partially decide
themselves when they
want to receive the
information
adequate, distract driver
not directly. Driver can
decide himself if and
when he wants to
receive information
hardly any literature, short
breaking impulses could
lead to an erroneous
interpretation of the driver
pedal-Feedback or
vibration on the steering
wheel helps the driver to
adopt to the speed limit
without nerving [3]
possible, but not
optimal, driver is
distracted from the
driving situation
hardly any literature
available
For the first level, the warnings, it become clear that acoustic signals (with the
exception of speech) are the most appropriate ones. The driver can handle the
signal instinctively. Visual as well as haptic signals are not adequate for this
priority level, because they are either interpreted too slowly or lead to
misunderstandings [4].
For the medium priority level the visual information is seen as the best solution,
because messages of this type are not too penetrating (like acoustical signals).
Haptic feedback, like it is used for example for ISA (Intelligent Speed
34
Drivers’ distraction due to ITS use
Adaptation) systems, is also seen as adequate. If the driver is speeding above
the limit the accelerator pedal gives a counter pressure which informs the driver
about his/her erroneous behaviour [5].
For the general information on the third priority level it looks as if the visual
information were the most appropriate one. The reason for this is that the driver
could decide him/herself when he/she wants to receive the information. Also
acoustic signals are adequate, but have to be announced in some way in order
not to distract the driver from the driving task. No literature was found for the
haptic signals on this level.
Based on the matrix above the following hypotheses for the systems which
were tested were formulated:
•
If level 1 warnings are submitted acoustically, than they have less negative
effect on the driving behaviour than if they are submitted visually.
•
Latent instructions of level 2 have less negative effect on the driving
behaviour if they are submitted via the haptic channel than if they are
submitted acoustically.
•
General information of level 3 has less negative influence on the driving
behaviour if it is given visually than if it is submitted acoustically.
In order to test these general hypotheses five scenarios, which represent typical
situation in real traffic, were developed. For these scenarios the driving task
was defined as "primary task" while the secondary task consisted of the tasks
given by the different telematic systems. In every scenario two systems
submitted information/warnings/instructions to the test persons. The task of the
test persons was to react correctly to these information/warnings/instructions. In
order to test the hypotheses the test persons drove through the test track twice
while the modality in which the information in the different scenarios were
submitted changed.
In the following, an overview about the scenarios and the systems which were
used.
35
Human Centred Design for Intelligent Transport Systems
Table 2. Overview of the experimental design of the simulation test
Scenario
1
Route Guidance
Information priority level 2
Traffic news
Information priority level 3
Scenario
2
ISA
Information priority level 1-2
Traffic news
Information priority level 3
Scenario
3
Parking information
Information priority level 2
Additional parking
information
Information priority level 3
Scenario
4
Pedestrian warning
Information priority level 1
Route Guidance
Information priority level 2
Scenario
5
Distance warning
Information priority level 1-2
Traffic news
Information priority level 3
Course A
Test situation
Control
1&2
Situation
Course B
Test situation
Control
1&2
Situation
acoustic
traffic sign
haptic
traffic sign
acoustic
-
acoustic
-
haptic
traffic sign
acoustic /
visual
traffic sign
acoustic
acoustic
acoustic
acoustic
acoustic
traffic sign
visual
traffic sign
visual
-
acoustic
-
visual
-
acoustic
-
acoustic
traffic sign
haptic
traffic sign
visual
traffic sign
acoustic
traffic sign
acoustic
-
acoustic
-
3 Simulator set-up
In order to give answers to the formulated research questions and hypotheses a
driving simulator based on a 3D development environment (BLENDER – open
source) was developed.
The test track consisted of roads within build up area and rural roads. Traffic
signs were virtualised in order to guide the test person through the course, and
speed limits as well as overtaking bans were announced. The general speed
limit was 50 km/h but on certain parts of the test track there was a 30 km/h limit.
Random generated traffic volumes but also virtualised cars and pedestrians
based on the behaviour of the test person appeared during the test rides.
The duration of one drive through the whole test track was about 15 minutes.
Altogether, including a five to ten minutes adaptation phase before each test
drive, the duration of the test was about 45 to 50 minutes.
A customary steering wheel for racing games was used as well as two pedals
(acceleration and breaking pedal as in a car with automatic gearing). A simple
engine noise reflecting the driven speed was heard.
The picture of the test track was projected with two beamers on a white wall.
The test person sat approximately two meters away from the wall.
36
Drivers’ distraction due to ITS use
Data about deviation from the optimum driving line, control of pedals (speed)
and steering wheel movements (deviation from the zero position) were logged
during the whole test ride. In addition, all test rides were filmed with two video
cameras.
The haptic information was given by two vibrating electro-motors which were
put on the left and right side of the test persons seat. For the haptic rout
guidance only the motor on the side in which the test person should go vibrated,
for the ISA warning both motors were vibrating when the test person drove
above the allowed speed limit.
4 Evaluation
The sample was distributed in the following way. Eleven men and eight women
participated in the simulator tests: The test persons had a mean age of 37,5
years with a standard deviation of 11,8:
Table 3. Sampling distribution
Under 30 years
between 30 and 49 years
50 years and above
Total
Male
4
5
Female
3
3
Total
7
8
2
11
2
8
4
19
The yearly driving performance showed that eight test persons drove less than
5.000 kilometres the year before, six drove between 5.000 and 10.000
kilometres, three between 10.000 and 20.000 and finally two test persons had a
yearly driving performance of more than 20.000 kilometres.
Four test persons stated that they had experience with computer racing games
while 15 test persons answered that they did have not. Seven test persons did
not have any experience with the warning and information systems used in the
simulator study while twelve participants said that they already had used one of
the systems at least sometimes.
Two evaluation methods were applied:
4.1
Statistical evaluation for speed and steering wheel
movements
In order to make the data of the test persons and for each situation comparable
only the data logged ten seconds before and after each scenario (test situation
and control situation) was taken for the statistical evaluation. Mean speeds for
the various situations and standard deviations of the steering wheel movements
were used to calculate significance's between the control situation and the two
test situations and within the test situations (Mann Whitney U-Test).
37
Human Centred Design for Intelligent Transport Systems
4.2
Behaviour observation with the help of the video
recording
In order to evaluate differences in the behaviour of the test persons in each
situation the recorded material was seen trough with the help of a standardised
observation sheets. In these sheets specific criteria were defined which cannot
be identified in the logged data,. e.g. overtaking where the test persons were
not allowed to overtake, conflicts with other road users, etc. The observed
driving behaviour in each test situation was compared between the control
situation and the two test situations and within the test situations. The aim was
to evaluate possible differences of the effect of the modalities in which the
information/warning was given, on the driving behaviour. Furthermore, on the
three parking places the test subjects were asked about the traffic news which
they received during the test- and control-situations, and answers were also
evaluated.
5 Results
In the following some basic results for each scenario are presented:
Scenario 1: The driven speed was significantly higher in the first test situation in
comparison to the control situation (46,73 km/h to 34,83 km/h, p=0,026). Also
the standard deviation of the steering wheel movement in both test situations
(acoustic route guidance/acoustic traffic information & haptic route
guidance/acoustic traffic information) was significantly higher than the one in the
respective control situations (32,71 to 26,22, p=0,000 & 32,25 to 24,90,
p=0,008), Additionally the behaviour observation showed that with haptic route
guidance three test persons did not follow the correct route.
Scenario 2: The standard deviation of the steering wheel movement was
significantly stronger in the situation were acoustic/visual ISA-warning was used
in combination with simultaneous acoustic traffic news than in situations were a
haptic ISA system was tested in combination with simultaneous acoustic traffic
news (17,2 to 14,9, p=0,011). The behaviour observation showed no
differences in the driving behaviour between the two test drives.
Scenario 3: No significant differences in speed and standard deviation of the
steering wheel movement between the two test drives. More erroneous turns on
the parking place were observed while using the visual route guidance in
combination with the additional acoustic information where to park, in
comparison with the situation in which an acoustic route guidance was tested in
combination with additional visual information.
Scenario 4: The standard deviation of the steering wheel movement was
significantly lower in the situation were acoustic pedestrian warning was tested
in combination with simultaneous haptic route guidance in comparison to the
situation where a visual warning system was used in combination with
simultaneous acoustic route guidance (29,7 to 32,1, p=0,040). The behaviour
observation showed that the visual warning was detected later and that it was
also more often ignored by the test persons.
38
Drivers’ distraction due to ITS use
Scenario 5: The standard deviation of the steering wheel movement was
significantly higher in the situation where an acoustic distance warning system
was tested in combination with simultaneous acoustic traffic news in
comparison to the situation were a visual distance warning system was used in
combination with acoustic traffic news (22,9 to 20,7, p=0,045). The behaviour
observation showed that the participants ignored the acoustic warning more
often than the visual one.
The results of hypotheses tests were transferred back to the matrix in which the
information given on different priority levels was combined with the three
modalities. The following table gives an overview about the comparison
between the literature results and the results of the simulator study.
Table 4. Classification of information into priority levels and modalities
modality
priority levels
acoustic
visual
haptic
warnings, high
priority
Confirmed by the results
of the simulator study
Confirmed by the
results of the
simulator study
Was not tested within
these project
medium priority
Confirmed by the results
of the simulator study
Confirmed by the
results of the
simulator study
Confirmed by the results of
the simulator study
general
information, low
priority
Confirmed by the results Was not tested within
of the simulator study.
these project
Results of the simulator
study are not clear in this
respect
Additionally, general results show that
•
None of the test persons could accomplish all tasks or reacted correctly to
all given information
•
None of the test persons could answer all question concerning the traffic
news correctly
This underlined that simultaneous communication should be avoided whenever
possible, as it may lead to an overload situation for the driver. Hence, it is
recommended to delay the communication of less important information until a
higher-priority warning has been resolved.
6 Workshop and resulting guidelines
The results were presented to Austrian telematic experts during a one-day
workshop. Additionally to the comments which were collected with respect to
the results also feedback about the matrix, future trends and recommendations
for the formulation of the guidelines were obtained.
Finally, the results from the literature study at the beginning of the project, the
results of the simulator study and the recommendations produced by the
experts were used to formulate guidelines for the optimum use of in-vehicle
telematic systems. These guidelines were compared to international guidelines
[6-8] so that at the end several LIVES-recommendations were the outcome of
the project:
39
Human Centred Design for Intelligent Transport Systems
•
Information and warnings have to be as simple as possible; they should
include clear instructions for the drivers, what kind of actions have to be
taken; it has to be clear for the driver what he/she has to do
Table 5. Recommendations classified into priority levels and modalities
modality
priority level
warnings,
high
priority
acoustic
sounds earcons and
2
analogue signals , but no
speech
medium
priority
general
information,
low priority
sounds and speech (also
if much information has to
be submitted to the driver)
visual
haptic
can be exceptionally used;
for instance related to
exceeding of limits (speed,
distance to the car ahead)
user should train the
symbols and pictures in the
handling of the haptic
direct visual field of the
signal before using it in
driver
real traffic
symbols, pictures, text and
maps; information should
be understood within two
seconds
Submitting simultaneous information:
•
There should be the possibility that priority level 3 information can be turned
off in order not to interfere with information of higher priority.
•
The simultaneous submission of information to the driver should, if possible,
generally be avoided. The possibility to receive information from two
different sources is limited and distract the driver from the driving task.
•
For certain priority level 3 information speech could be used. But one
should not submit more messages from this area simultaneously. Priorities
have to be set, submission should happen consecutively.
•
Information with lower priority should be submitted in such a way that
information/warnings of higher level priority can be understood without any
problems.
It is possible to submit the information with the same content multi-modal so
that two or more modalities are used and strengthen the information. This can
reduce the reaction time of the driver.
•
The use of different modalities, especially acoustic and visual, for the
simultaneous submission of the same information is possible without
disadvantages.
2
Sounds - noises, tones: random mapping, the signal has no direct link to the content of
the information
Earcons - mnemonic mapping, the signal has a semantic link to the content of the
information
Analogue signals -analogue mapping, the signal is according to the information modified
along a dimension
40
Drivers’ distraction due to ITS use
•
The use of acoustic and visual information in combination with haptic
submission of the same information is possible for priority level 2 warnings.
There, the acoustic and visual signals support the understanding of the
haptic signal
7 Conclusion
The aim of this project was on the one hand to give recommendations for which
modality should be used to submit information to the driver while considering
the priority level of the content. On the other hand, the goal was to look how
simultaneous information submitted to the driver influences his/her driving
behaviour. This was done in the light of the fact that more and more in-vehicle
telematic systems are today implemented in cars in order to inform, warn and
help the driver. Conclusions were drawn as a result of simulator tests as well as
from comments by experts in different settings.
Research work in this project provided indications for what modalities should be
used for different kinds of information and warnings in order to address the
driver in the best way, without interfering too much with the driving task.
Acoustic information can be used when low priority information is submitted, as
it is usually not too much disturbing. However, this modality also should be used
when the driver should (urgently) be warned, because information forwarded via
this channel is interpreted very fast. Visual information should mainly be used
for low and medium priority information. Like for the acoustic modality it is
essential to consider how visual information should be prepared for the driver.
Low-priority information can be presented in detail (like maps). The higher the
priority becomes the more simple the information should become (symbols).
The haptic modality can with advantage be used for the medium priority level.
Simultaneous submission of information should be avoided unless two different
modalities are used to submit the same information. If different types of
information become relevant simultaneously a “workload manager” or “dialogue
manager” could help, that "decides" what information should be presented to
the driver first and what information could be suppressed for the moment and
provided later, under better preconditions. Although such technical solutions are
much sought for, their development must be based on socio-scientific findings,
such as the guidelines described in this project. With the help of this research,
solutions for both individual systems and combinations of systems can be
developed.
The clear advantages in using a simulator was that with the help of this
approach many different information and warning systems could be tested
(which would be rather complicated in real traffic). Especially for the warnings of
priority level 1 the controlled setting of the simulator would not have been
achieved in real traffic. The approach allowed to gain basic results concerning
the research questions in a cheap and easy way. The sample size was rather
small but the statistical evaluation nevertheless showed significant differences
between the used modalities and systems. Moreover, the results of the
simulation study were discussed with experts and compared with the actual
literature in order to validate both the outcomes and the final product, namely
the guidelines.
41
Human Centred Design for Intelligent Transport Systems
However, more research within this field has to be done. For instance, there is
hardly any literature according to haptic information which may be seen as a
good possibility for in-vehicle telematics, as this channel is not used so far.
More information also has to be gathered about technical solutions to avoid
simultaneous submission of information (“dialogue manager”). Furthermore,
recommendations and directives for HMI application should be continuously
adjusted. Not least, clear instructions for the use of telematic systems, and for
training for their use, should be given.
8 Acknowledgement
This project was funded by the project line I2 – Intelligent Infrastructure of the
Austrian Federal Ministry of Transport, Innovation and Technology (BMVIT) in
co-operation with Österreichische Forschungsförderungsgesellschaft mbH
(FFG),
9 References
[1]
[2]
[3]
[4]
[5]
[6]
[7]
[8]
internal project report: Geven A., Kaufmann C., Risser R., Schrammel J.,
Sefelin R.: LIVES -LenkerInnenInteraktion mit VErkehrstelematischen
Systemen (FACTUM OHG, Vienna, 2006)
Wickens, C.D., Gordon, S.E., & Liu, Y.: An introduction to human factors
engineering (New York: Addison-Wesley Educational Publishers Inc.,
1998)
Lee, J.; Stoner, H. & Marshall, D.: Enhancing interaction with the driving
ecology through haptic interfaces. IEEE International Conference on
Systems, Man and Cybernetics, 2004, 1, 841-846.
Rombaut, M.: Prolab 2: A driver assistance system. (Mathematical
Computer Modelling, 22(4-7), 1995), pp.103-118
Kaufmann, C. & Risser R.: Die Effekte der Verwendung eines aktiven
Gaspedals AAP in einem Feldversuch in Lund (Zeitschrift für
Verkehrssicherheit 4/2003, 2003) pp. 184-190.
2000/53/EG: Empfehlung der Kommission vom 21. Dezember 1999 an die
Mitgliedstaaten und die Industrie über sichere und effiziente On-boardInformationsund
-Kommunikationssysteme:
Europäischer
Grundsatzkatalog zur Mensch-Maschine-Schnittstelle (Bekannt gegeben
unter Aktenzeichen K(1999) 4786).
Japan Automobile Manufacturers Association Guideline for In-vehicle
Display Systems — (Version 3.0., 2004)
Green, P., Levison, W., Paelke, G., Serafin, C. (1994). Suggested Human
Factors Design Guidelines for Driver Information Systems, The University
of Michigan Transportation Research Institute (UMTRI)
42
Drivers’ distraction due to ITS use
NATURALISTIC DRIVING OBSERVATIONS TO
INVESTIGATE DISTRACTION EXPOSURE AND IVIS
PATTERNS OF USE: INTERESTS AND
CONSTRAINTS OF THE APPROACH
Arnaud Bonnard, Corinne Brusque (INRETS, France)
arnaud.bonnard@inrets.fr
ABSTRACT: With the development of communication and information
technologies, driver’s distraction linked to In-Vehicle Information Systems has
become a key issue for road safety. Despite ten years of investigation on IVIS
impacts on driving behaviour, several research issues remain to be addressed.
By observing in an unobtrusive way the behaviour of drivers, at the wheel of
their own vehicle, during their daily journeys, naturalistic driving observations
seem an interesting approach to investigate the link between driver distraction
and critical situations occurrence, driver exposure to distraction related or not to
IVIS and IVIS patterns of use. Despite the current technological offer permits
their implementation, these studies raise some constraints, in terms of ethical
issues, participants’ selection, servicing processes, data coding and statistical
analysis. Nevertheless, even if the issues to address are various, the outcomes
of naturalistic driving observations in term of knowledge about IVIS integration
in everyday life driving deserve the challenge.
1 Introduction
With the development of communication and information technologies and their
deployment in the field of road transport, In-Vehicle Information Systems (IVIS)
is becoming a common feature of the car environment. IVIS provide various
functions and services to the driver. They may deliver information related to the
trip management (e.g. navigation, traffic or weather information) but also
completely unrelated to the driving activity (e.g. phone calls, e-mails…).
However, driver’s distraction linked to the use of these electronic devices has
become a key issue for road safety. Indeed, even if it is expected that IVIS can
increase driving comfort when supporting the driver adequately in his or her
driving task, it has been demonstrated that IVIS handling or checking can have
negative side effects, such as distraction, mental workload increase, lapses of
attention, reaction delays and inadequate situation awareness.
Since the nineties, numerous researches have been carried out, using various
experimental contexts (driving simulators, test tracks or real traffic conditions) to
estimate driver’s demand while performing this kind of secondary task and to
evaluate how this workload increase interferes on the driving performances [13]. In parallel, some authors showed that drivers engaged in secondary tasks
have an increase risk of being involved in a crash or near-crash, whatever the
solicitation modality (auditory, visual or manual) [4-6]. Accident analysis
databases have also been analyzed to identify the part of distraction on crashes
[7-8]. Aware of this issue, the European Commission, through the edition of
European Statement of Principles on the design of Human Machine Interaction,
decided to disseminate widely the best practices for the design of IVIS and their
43
Human Centred Design for Intelligent Transport Systems
in-vehicle installation [9-10]. Their objective was to limit the risk that poorly
designed interfaces introduce unnecessary driver’s distraction. From a few
years, driver’s distraction due to IVIS use has become a major topic of interest
for researchers, authorities and designers. Despite this interest and the efforts
achieved, some issues remain unresolved.
We assume that naturalistic driving observations offer the opportunity to fill in
these research gaps. Indeed, it will be a challenging approach to investigate
drivers’ distraction exposure and patterns of use of IVIS in everyday driving. In
this paper, we present the main characteristics of naturalistic driving
observations. Then, we highlight the research issues they can address and we
describe the observations that can be carried out with this approach according
to the current technological supply. However, the interests of these new
possibilities of investigation must be balanced with their constraints. Thus, the
different methodological questions raised by this approach are presented and
discussed.
2 Naturalistic driving observations
2.1
Definition
Naturalistic driving studies consist in observing, in an unobtrusive way, drivers
behaviour in naturalistic settings - that is, during their everyday life driving - out
of experimental context [11-13]. Indeed, participants just drive where and when
they want to, at the wheel of their own car. Participant cars are equipped with
an instrumentation package, in order to record different information. Thus, it is
possible to collect data related to the driver’ state, to his or her driving
behaviour, to the vehicle dynamics, to the infrastructure or to the environment
surrounding the car. These data are recorded non-stop during all the
observation period. After the observation, they permit an in-depth analysis of
drivers’ behaviour by researchers. Recorded and analyzed drivers’ behaviours
can concerns standard driving situations but also critical or accidental
situations.
Thus, conducting naturalistic driving observations seems a relevant method to
collect information on the way drivers use IVIS in everyday driving, in addition to
traditional crash investigation or experimental studies.
2.2
Research issues that can be addressed
Despite ten years of investigation on drivers’ distraction and IVIS impacts on
driving behaviour, several major issues of research remain to be addressed.
However, this missing knowledge can be largely investigated by naturalistic
driving observations under the three following aspects at least: the link between
driver distraction and critical situations occurrence, the drivers’ exposure to
distraction related or not to IVIS and the patterns of use of IVIS.
Researchers that attempt to investigate the part of distraction on road accidents
are confronted with the weakness of police reports and crash investigations
regarding information on driver’s distraction or inattention at the moment of the
crash [7-8]. Indeed, this distraction or inattention information is only issued from
self-report of drivers involved in crash, or eyewitness accounts. But drivers
44
Drivers’ distraction due to ITS use
generally try to hide some details of what really occurred for fear of prosecution.
Furthermore, it is difficult for the crash eyewitness to establish if the driver was
running a secondary task just before the accident. At last, the recording of
distraction or inattention information is not currently generalized in all accident
databases. For example, in spite of the high media coverage of the risk linked to
mobile phone use while driving, this information is not systemically available.
And the solution retained by Redelmeier [4] to match the crash information
given by the police reports to the phone activity of the drivers few minutes
before the crash given by their telecommunication company is also spoilt by
mistakes due to the inaccuracy of the collision time. For these reasons,
naturalistic driving observations offer an efficient way to evaluate objectively the
part of distraction on accidents with a possible in-depth analysis of crash and
near-crash events recorded during the observation period [6].
Moreover, there is a lack of knowledge concerning the driver exposure to
distraction. Authorities focus on the distraction risk due to electronic devices
handling or checking, but the relative part of this specific kind of distraction
amongst all the other potential distraction sources needs to be investigated. For
example, what are the driver’s exposures to adjusting ordinary vehicle controls
as climate control or radio, to locating or reaching objects inside the vehicle, to
being distracted by car passengers, to being distracted by billboards along
roads or to being lost in thoughts? And what is the actual exposure to handling
or checking IVIS? If questionnaire-based surveys permit to distinguish clearly
drivers that decide to use IVIS or not while driving, they are less relevant to
evaluate accurately the frequency of IVIS use and the characteristic of the
drivers that are more likely to use them. This information is necessary for an
accurate exposure analysis and can be provided by naturalistic driving
observations [11].
Beyond frequency of IVIS use, it is important to identify the patterns of use of
IVIS in everyday driving. Under experimental conditions, subjects are instructed
to pay maximum attention to the driving task, but they are still in a position to
run a secondary task at the experimenter’ request. We assume that in the
everyday driving, some drivers decide to minimize the risk due to electronic
devices handling or checking by choosing carefully the driving context for their
use. For example, they can decide to perform some interactions only when their
vehicle is stopped. They can also decide to limit IVIS use according to the type
of road or the traffic condition. They can also defer some interactions waiting for
a more suitable driving context. Finally, they can slow down in order to keep
safety margin thresholds at an acceptable level. Thus, it is interesting to
highlight the detailed modalities of driver interactions with IVIS. Furthermore,
IVIS are designed for a reference use that is defined by the objectives of the
systems and the conditions in which it should be operated. We assume that
some drivers might misuse their systems, using them for objectives or in
conditions that were not anticipated by their designers. For example, a driver
could activate the cruise control system in order to be more available to read
the SMS message he has just received on his cell phone. This unexpected uses
of systems, called misuses, have to be evaluated and quantified. Finally, when
providing drivers with driving-related information, IVIS can modify driving
behaviour at strategic, tactical or operational levels. The way drivers really use
all the available IVIS information, the related modifications of their travel pattern
and the long term effects on their driving behaviour are still unknown. These
45
Human Centred Design for Intelligent Transport Systems
three different aspects of patterns of IVIS use could be evaluated thanks to
naturalistic driving observations.
Some North-American researches attempted to answer these questions [11-13].
Nevertheless, their outcomes can not be applied straight away to the European
drivers because of major differences on vehicle models, on road network
characteristics and on available technologies on the market. European
countries also present cultural, institutional and economic differences that can
explain various levels of IVIS availability or different drivers’ willingness to use
IVIS. All this argues to launch naturalistic driving observations in Europe in
order to address all these unresolved research issues, all the more so since the
technological offer permits nowadays their setting-up on a large scale.
2.3
Available technology
Nowadays, actual technological equipments enable us to easily and efficiently
equip the participants’ personal car and to collect several hours of data and
video. Basically, a simple video recording system would be a good basis to
study naturalistic driving behaviours, by filming what happens inside and around
the vehicle [11]. However, a broad range of sensors can come in addition to this
system [13]. Various vehicle parameters, like speed, steering wheel angle or
indicator use can be collected via the E-OBD interface, by scanning the CAN
bus of the vehicle or via specific additional sensors. The instrumentation
package could also include a GPS to collect information on the vehicle
whereabouts or several accelerometers to collect the lateral and longitudinal
acceleration of the vehicle. We can even imagine the integration of advanced
perception sensors like lane tracking systems or radars to know the position of
the car on the road, and the surrounding traffic. However, these sensors are still
expensive and their relevance is yet to be determined, as some of this
information can be computed a posteriori from the video with image processing
algorithms. Furthermore, given the constantly increasing size of the hard disk
drives in computers, such a hardware data collection system could collect hours
and hours of continuous driving data.
Obviously, it is easily possible to use many different sensors for the data
collection with the only constraints of cost and of available space. However, to
perform an in-depth investigation of driving behaviours, the recorded
parameters will have to be carefully analysed.
To ease this analysis, the large amount of collected data can be automatically
processed with the objective to find models that may deliver interesting
behavioural information. For example, triggers can be set in order to select the
near-accident sequences [13] or other potentially interesting sequences. To do
so, the recorded parameters must include data that might act as triggers to
indicate interesting sequences (e.g. beginning of a phone call, entering of a
destination…). Even if the relevance of these sequences still has to be validated
by an expert, it will facilitate and optimize the analysis work. However, some
behavioural information can only be found manually by experts. This implies
that they will have to check cautiously most of the collected data and videos, in
order to code relevant behavioural information, which is very time-consuming.
Finally, technological choices must be a compromise. A simple data collection
system offers a higher reliability and will be simpler and faster to equip in the
46
Drivers’ distraction due to ITS use
participants’ cars. However, due to the complexity and the cost of naturalistic
driving observations, the instrumentation should still be rich and open enough to
allow future exploitations.
Even if naturalistic driving observations are promising in terms of research
issues to address and in terms of instrumentation development, the
implementation of this kind of studies raises a set of additional constraints.
3 Constraints of Naturalistic driving observations
3.1
Ethical issues
First of all, from the ethical point of view, it is important to respect the privacy of
the drivers and of the potential passengers. Indeed, data protection is a
fundamental human right that must be guaranteed by experimenters, even if
data protection laws may differ from a country to another.
Thus, experimenters must ask participants to sign an informed consent form.
This consent form explains the purpose and the constraints of the study. It also
explains that the confidentiality of the personal data is guaranteed and that their
personal data will be anonymous. Finally, it offers them the possibility to
withdraw from the study at anytime. When signing the consent form,
participants fully agree to be involved in this study and that their personal data
will be analysed by experimenters for the research needs. The difficulty is that
the instrumented vehicle can be driven by secondary drivers and drivers can
also carry some passengers, adults but also children and teenagers.
Experimenters could then have difficulties for collecting consent forms of these
casual secondary drivers or passengers.
Moreover, the respect of personal data protection can have a major impact on
the kind of data that can be recorded during naturalistic driving observations,
and especially concerning the video and audio recording systems. Different
technical solutions can address this issue. The inside camera could only
capture the driver’s face and not the entire vehicle cabin to guarantee the
anonymity of potential passengers. Also, the audio information could not be
recorded with the video to respect participants’ private life [13]. This kind of
restrictions was used during previous studies, even if it considerably decreases
the quality of dual task situations analysis. Indeed, with such recording
conditions, it is hard to distinguish if the driver is talking with a passenger or if
he/she conversing on a hands-free phone. It would also be difficult to identify
which IVIS is handled by the driver [13].
To guarantee an accurate analysis of dual task situations, it seems preferable
not to restrict the collected parameters, but to allow the participants to stop the
data recording when they want to, with the only constraint to inform the
experimenters of the reason of their decision. Thus, when participants take part
in long observation period, they are able to preserve their private life when
exceptionally needed, with only little impact on the final analysis.
47
Human Centred Design for Intelligent Transport Systems
3.2
Definition & representativeness of the driver’s sample
The correct definition of the sample of drivers involved in a naturalistic driving
study and the capacity to generalize its results to the entire population are also
major issues.
For experimental studies, the subjects are usually selected according to their
gender and their age class. But in the case of a naturalistic driving study
investigating the patterns of IVIS use, the participants’ sample should not be
only constituted according to these two criteria [11] and other parameters
should be taken into account.
In order to avoid any biased results, it is necessary to check if the devices at the
disposal of the participants are sufficiently representative of the market. Indeed,
the commercial offer for IVIS is diversified between the OEM systems, the after
market ones and the nomadic devices. The human machine interfaces of the
systems are also varied. They can be based on a visual/manual or on a vocal
interface; the navigation in the menus can be more or less complex and so on.
As these interaction modalities can have a direct and strong effect on the
impact of IVIS on drivers’ behaviour, they have to be selected very carefully.
Of course, the fact that drivers are regular or irregular IVIS users must also be
considered. In order to detect significant associations between IVIS use and
driver’s behaviours, and even more to detect associations between IVIS use
and crash or pre-crash situations, it is necessary to collect various patterns of
IVIS use in sufficient quantities. However, even if it is practically preferable to
use a sample with intensive IVIS users in order to clearly highlight typical
patterns of use and to limit the necessary observation period, the possibility to
generalise the results should be balanced. Indeed, drivers’ population is
composed of individuals with very heterogeneous IVIS handling habits and
irregular users also have to be taken into account.
Finally, the sample has also to take into account the various travel patterns of
the population in order to highlight different patterns of use of IVIS and different
conditions of distraction exposure. For example, concerning distraction, we
assume that a commuter will probably be easily distracted by a phone call but
maybe won’t be aware of outside distractions like special road signs. On the
opposite, a couple with children that mainly drive to the countryside on the week
ends will be more likely distracted by a navigation system (or by their children!).
It will be necessary to control the participants’ sample in terms of numbers of
driven kilometres, of road networks taken and of travels purposes.
As a conclusion, it is important to control the characteristics of the sample, not
only in terms of gender and age. Even if we try to obtain a representative
sample, a correction of the exposure data will still be necessary to extend
accurately the results to the entire population.
3.3
Suitable servicing processes setup
Another major issue is to guarantee the quality and the consistency of the data
collected during the whole observation period. First of all, this can be achieved
by focusing on the simplicity and on the robustness of the data recording
system [13], as it is a good solution to prevent potential hardware or software
failures. However, it is also important to have additional processes to detect any
48
Drivers’ distraction due to ITS use
possible breakdown. It is necessary to organise suitable servicing processes.
This servicing will have to be reactive and mobile, given that the car can be
anywhere. It is also important to plan repetitive partial data downloads to
minimize the potential risks of data loss [13]. All these processes will permit to
fix most of the problems. It will guarantee a maximum quality of the information
collected during the whole experimentation. However, to limit the workload and
the journeys of the servicing team, a periodic remote monitoring of functioning
of the technical equipments installed in the participants’ car can be assured. For
example, the system could send automatic reports via a wireless connection
about the functioning status of its different components, but also summary
information about the last collected data [14]. Moreover, download could also be
done remotely using a wireless connection. Communication strategies will be
necessary to keep a connection with the car even when it goes out of network
coverage.
3.4
Event selection and data coding issue
As it was briefly presented in part 2.3, the selection of the significant events for
the statistical analysis and the additional coding of the selected sequences are
crucial issues to obtain relevant results from the naturalistic driving observation.
Indeed, according to the events studied by researchers (phone calls, consulting
traffic or navigation information, pre-crashes and incidents) and their occurring
frequency while driving, it is necessary to observe drivers and to collect data
continuously during more or less long periods (a few weeks, several months or
even a year). Amongst this huge amount of collected data, only a small subset
of sequences is significant for the statistical analysis. Different kinds of
procedures can be used to select these sequences. First, events can consist of
few-seconds or few-minutes period of time selected randomly. This procedure
has been used to evaluate drivers’ distractions exposure [11-12]. Events can
also be selected from event triggers. For example, for the identification of precrashes or incidents, a set of variables measuring vehicle kinematics with a set
of associated trigger criteria have been used to pre-select applicant events,
selections are later strengthened by watching the corresponding video clip
[13][15]. We can also dream up to conceive event triggers that identify the
moments when drivers are using their electronic devices, with an analysis of
video or audio data, or with specific sensors of devices running. This latter
solution will be easier to set up when the instrumented vehicles are lent to
subjects and then experimenters can control IVIS present in the vehicle.
All in all, this reduction process is necessary to reduce the effort for the
following step, the data coding. Indeed, this part of the study is very timeconsuming, but the accuracy of this work has a direct impact on the quality of
the outcomes of the research. Video data plays a major part in the data coding
process. Video clips permit first to evaluate if the driver is attentive or distracted
and in the case of distraction to evaluate the distraction source. It permits also
to code information about the driving environment like road characteristics,
traffic, light conditions and so on. At last, the driving task under way can only be
described in detail by watching video. Moreover, the coding of video data has a
subjective nature and because of the amount of data, several people could be
involved in this activity during several months. For that, they must be trained,
49
Human Centred Design for Intelligent Transport Systems
and the quality of their must be controlled on a regular basis with the objective
to increase inter and intra reliability [13].
3.5
Statistical analysis constraints
The last major issue raised by naturalistic driving observations concerns the
statistical analyses and their constraints that differ according to the research
objectives. For example, for researches that attempt to evaluate drivers’
distraction exposure, the challenge is to extend the results obtained for the
subjects’ sample to the whole population. We have already tackled the
constraints raised by such an analysis in section 3.2, so we don’t broach this
topic again and we prefer to focus on the case of researches attempting to
evaluate the impact of IVIS on driver’s behaviour.
Statistical approaches have already been tested to analyse data collected
during naturalistic driving studies. However, they have to be applied carefully.
For example, an approach consists in selecting random driving sequences in
order to assess drivers’ distraction status, according to the fact that the driver is
distracted or not, and according to the nature of the distraction [12]. Then, these
authors evaluate the impact of distraction on drivers’ behaviour by aggregating
all the sequences with the same distraction status and by looking for significant
differences on driving performance among them. However, this aggregating
process doesn’t into account the heterogeneous driving context of the selected
sequences, which can mask specific dual task effects.
Indeed, in the case of naturalistic observation, unlike experimental research, all
the situations of dual task studied for each participant are unique, in terms of
secondary task nature and also in terms of driving context. Here, it is not
possible to isolate the effect of dual task by controlling all the interfering
variables by the experimental design. A possible solution would be to realize
matched-paired comparison. For that, each driver is considered as his/her own
control and each situation of dual task must be matched with a baseline
situation. The baseline situation must be as similar as possible to the dual task
situation in terms of driving context. For example, a dual task situation can be
matched with a sequence of the same duration, randomly selected just before,
or just after, the studied sequence and during which the driver must be not
distracted by any source of distraction. It is also possible to identify sequences
with similar characteristics, like manoeuvre, traffic condition and road
infrastructure, using advanced mathematical modelling like hidden Markov
modelling [16]. This new approach would bring interesting information about the
impact of IVIS use while driving.
Nevertheless, the effect of IVIS handling and checking on drivers' behaviours at
operational level must be certainly hard to evaluate. Indeed, the experimental
studies on mobile phone use while driving show with a clear consensus that
performing secondary task increases reaction time [1][3]. Conversely, the
results concerning impacts on vehicle control (e.g. speed, headway, lateral
position) are divergent [1][3]. And even for the authors who found significant
effects, these effects are low. Thus, for naturalistic driving observations in
everyday driving, it could be difficult to show effects on vehicle controls
specifically due to the dual task situation and not to some variations in the
driving context. Furthermore, the setting-up of compensatory processes during
dual task situations is linked to the driving context, but it is also linked to
50
Drivers’ distraction due to ITS use
personal characteristics of drivers, like the awareness of the risks taken to
handle or to check an IVIS, or the driving style.
The statistical approach retained for the data analysis, and its specifics
constraints, must be taken into account at an early stage of the design of the
naturalistic driving observation in order to make sure to achieve the desired
research outcomes.
4 Conclusion
As a conclusion, even if naturalistic driving observations raise issues of different
natures, they are still an interesting challenge that promises many visionary
outcomes and that will certainly bring us new knowledge and new
methodological approaches. Very aware of this interest, European researchers
have already tackled this challenge and make strong efforts to implement
naturalistic driving observations, with the support of their government and of the
European Commission. We can assume that in a near future, large scale
naturalistic driving observations will be launched in Europe and will complete
the knowledge brought by American studies.
5 References
[1]
[2]
[3]
[4]
[5]
[6]
[7]
[8]
[9]
Caird, J. K., Scialfa, C. T., Ho, G., and Smiley, A.: 'Effects of cellular
telephones on driving behaviour and crash risk: results of meta-analysis'
(CAA Foundation for Traffic Safety, 2004)
Road Safety Committee: 'Inquiry into Driver Distraction' (Parliament of
Victoria, 2006)
Brusque, C. (Ed): 'The Influence of In-Vehicle Information Systems on
driver behaviour and road safety: Synthesis of existing knowledge', (Les
collections de l’Inrets - Synthèse Inrets, 2007)
Redelmeier, D. A., and Tibshirani, R. J.: 'Association between cellulartelephone calls and motor vehicle collisions', The New England journal of
medicine, 1997, 336, (7), pp. 453-458
McEvoy, S. P., Stevenson, M. R., McCartt, A. T., Woodward, M., Haworth,
C., Palamara, P., and Cercarelli, R.: 'Role of mobile phones in motor
vehicle crashes resulting in hospital attendance: a case-crossover study',
British Medical Journal, 2005, pp. bmj.38537.397512.55
Klauer, S. G., Dingus, T. A., Neale, V. L., Sudweeks, J. D., and Ramsey,
D. J.: 'The Impact of Driver Inattention On Near-Crash/Crash Risk: An
Analysis Using the 100-Car Naturalistic Driving Study Data' (Virginia Tech
Transportation Institute, 2006)
Stutts, J. C., Reinfurt, D. W., Staplin, L., and Rodgman, E. A.: 'The role of
driver distraction in traffic crashes' (University of North Carolina at Chapel
Hill - Highway Safety Research Center, 2001)
Stevens, A., and Minton, R.: 'In-vehicle distraction and fatal accidents in
England and Wales', Accident Analysis & Prevention, 2001, 33, (4), pp.
539-545
European Commission: 'Commission Recommendation of 21 December
1999 on safe and efficient in-vehicle information and communication
51
Human Centred Design for Intelligent Transport Systems
[10]
[11]
[12]
[13]
[14]
[15]
[16]
systems: A European statement of principles on human machine interface
(notified under document number C(1999) 4786)', Official Journal, 2000,
L019, pp. 0064-0068
European Commission: 'Commission Recommendation of 22 December
2006 on safe and efficient in-vehicle information and communication
systems: Update of the European Statement of Principles on human
machine interface (notified under document number C(2006) 7125 final)',
Official Journal, 2007, L 32, pp. 200-241
Stutts, J., Feaganes, J., Reinfurt, D., Rodgman, E., Hamlett, C., Gish, K.,
and Staplin, L.: 'Driver's exposure to distractions in their natural driving
environment', Accident Analysis & Prevention, 2005, 37, (6), pp. 10931101
Sayer, J. R., Devonshire, J. M., and Flannagan, C. A.: 'The effects of
secondary tasks on naturalistic driving performance' (The University of
Michigan - Transportation Research Institute, 2005)
Dingus, T. A., Klauer, S. G., Neale, V. L., Petersen, A., Lee, S. E.,
Sudweeks, J., Perez, M. A., Hankey, J., Ramsey, D., Gupta, S., Bucher,
C., Doerzaph, Z. R., Jermeland, J., and Knipling, R. R.: 'The 100-Car
Naturalistic Driving Study Phase II – Results of the 100-Car Field
Experiment' (Virginia Tech Transportation Institute, 2006)
'Website of LAVIA project', http://heberge.lcpc.fr/lavia/index.html, accessed
Jan. 2008
Dingus, T. A., Neale, V. L., Klauer, S. G., Petersen, A. D., and Carroll, R.
J.: 'The development of a naturalistic data collection system to perform
critical incident analysis: An investigation of safety and fatigue issues in
long-haul trucking', Accident Analysis & Prevention, 2006, 38, (6), pp.
1127-1136
Dapzol, N.: 'Weight Semi Hidden Markov Model and driving situations
classification for driver behavior diagnostic', 4th international Driving
Symposium on Human Factors in Driver Assessment, Training and Vehicle
Design, Stevenson, Washington, USA, 9-12 July 2007, 84-90
52
SESSION 2:
TOOLS AND METHODOLOGIES
FOR SAFETY AND USABILITY
53
Tools and methodologies for safety and usability
DEVELOPMENT OF INNOVATIVE
METHODOLOGIES TO EVALUATE ITS SAFETY
AND USABILITY: HUMANIST TF E
Sally Cotter (TRL, United Kingdom),
Alan Stevens (TRL, United Kingdom),
Anke Popken (Chemnitz University of Technology, Germany),
Christhard Gelau (BASt, Germany)
scotter@trl.co.uk
ABSTRACT: During the HUMANIST project Task the aim of Task Force E (TF
E) was to develop innovative methodologies to evaluate ITS safety and
usability. A matrix of methods for the assessment of ADAS and IVIS was
developed and subsequently a proposed model for integration of these
methodologies was proposed. A number of methods for the assessment of
driver appropriation of ITS over time were also established and another model
was proposed for the integration of these methods into one integrated
methodology.
1 Introduction
In-vehicle information and communication systems (IVIS) are increasingly the
standard equipment of modern cars. Advanced Driver Assistance Systems
(ADAS) are also increasing, but the boundaries between the definition of IVIS
and ADAS systems are blurred. Despite the obvious benefits of IVIS and ADAS
there are also concerns that their use while driving may cause safety problems
due to distraction and increased workload. Thereby it is assumed that it
depends mainly on the user-friendly design of the Human-Machine-Interface
(HMI) if the use of IVIS is compatible with the primary task of driving. In Europe
recent efforts to develop and evaluate a catalogue of design goals for the invehicle HMI of IVIS, the so called “European Statement of Principles” (ESoP),
reflect this approach. Nevertheless, the character of the ESoP is generic, i.e.
defining no criteria to assess if the design goals have been achieved by a
certain HMI solution. From all this it follows that there is an urgent need for
methods to evaluate the effects of IVIS on driver workload and behaviour in
order to assess potential problems for traffic safety and to improve HMI design
[1]. It is also vital that researchers start to understand how drivers learn to use
IVIS and ADAS and how this affects their driving performance over time as they
become more familiar with particular systems.
One of the main objectives of HUMANIST Task Force E (TF E) was to
exchange through the network, the knowledge and experience of projects which
have applied or developed methodologies for the evaluation of on-board and
off-board ITS; assessing systems both in terms of safety and usability. The
Task force was required to consider the integration of various methods into
comprehensive methodologies and then to consider the investigation of driver
appropriation processes overtime.
55
Human Centred Design for Intelligent Transport Systems
2 Innovative methodologies to evaluate its safety and
usability
The TF took a decision early in the project that the development of a matrix of
methods would assist in the conception and confrontation of methods and
procedures for usability and safety evaluations of ITS. The purpose of the
matrix was to identify and categorise existing and proposed methods for
usability and safety evaluation of ITS. When assessing usability and safety,
different metrics (or measures) are used. These metrics are collected using
specific techniques and often require specific tools, such as cameras or
questionnaires. The metrics are collected in one or more physical environments.
The following measures were collated for each method identified:
•
Metric: Measure used to assess safety and usability of ITS e.g. brake
response time.
•
Technique: Details how the metric is determined e.g. deviation in distance
between the vehicle centre and the road centre line
•
Tool: Equipment needed to use the metric e.g. video camera, adjusted
speedometer
•
Environment: Environment in which the test is carried out. e.g. Instrumented
vehicle
•
Aspect of the system/human investigated: Describes how the metric
assesses ITS safety and/or usability
•
Type of data: Objective, subjective or observed (expert opinion)
•
Effectiveness: How useful the measure is
•
Practical issues: Issues to consider when using metric e.g. time, expense
•
Organisation: Organisation which submitted or had experience of the metric
The combination of metric/technique/tool/ environment is what is referred to as
a method. The Task force held two workshops to discuss different metrics and
the formation of a matrix. A summary of some presentations can be seen in
HUMANIST Deliverable D.2/E.2 [2]. To populate the matrix information was
collated through these workshops as well as through scientific literature and
experience of the HUMANIST partners. The measures were categories into
nine groups; lateral control measures; longitudinal control measures (speed,
vehicle following); steering wheel movement measures; eye tracking measures;
physiological measures; situation awareness measures; task orientated
measures; subjective mental workload measures; incident analysis measures.
Initially only IVIS were considered then later in the task force a final category of
ADAS measures was included in the matrix.
That matrix contains 130 methods. Partners were asked to indicate their
‘favourite’ method. Favourite meaning those most frequently used or most
useful to the HUMAIST network partners. There are 22 favourites identified and
the range of favourites is very broad, demonstrating the variety of methods
available and that the methods used depend heavily on the objectives of a
56
Tools and methodologies for safety and usability
study. The ‘favourites’ were categorised at 3 levels according to the number of
partners selecting the method (level 1 being the most popular favourites and
level 3 being the least popular favourite). There were two ‘level 1’ favourites:
lane standard deviation and minimum following distance (closest longitudinal
approach). The ‘Level 2’ favourites were: occlusion, glance frequency, glance
duration, brake response time, time headway and time to line crossing. The full
matrix can be found in the HUMANIST deliverable E.4 [3].
The matrix summarises important information and experience about each
method based on scientific literature and experience of the HUMANIST
partners. This information includes the type of data obtained (e.g. objective,
subjective, observational or expert opinion); the effectiveness of the method
expressed in terms of the validity, reliability and sensitivity; and practical issues
of application including time, cost and training requirements. Where possible,
the matrix provides scientific references and lists the HUMANIST partners with
experience or knowledge of the methods. From developing the matrix it is clear
that partners are doing substantial research into the impact of IVIS on driver
workload and behaviour and that a large variety of ITS assessment methods
are used.
3 ADAS and IVIS
An ‘In Vehicle Information System’ (IVIS) and an ‘Advanced Driver Assistance
System’ ADAS are both obviously defined as systems. All systems perform
functions, such as obtaining an address from the driver, warning of a collision
or adjusting headway. Additionally some systems may include information
functions, warning functions and assistance functions (and also in-built stability
functions such as ABS). IVIS focuses on informing while ADAS focuses on
warning and assisting. The table below demonstrates the issues associated
with each of the four functions, hence distinguishing the difference between
each of the functions. A detailed explanation and definition can be found in
HUMANIST Deliverable E.4 [3]. An integrated methodology needs to address all
of the safety issues listed in the diagram.
FUNCTION
In-built
Informing
Warning
Assisting
ISSUE
Example
ABS
ASC
Collision mitigation
Route guidance
Mobile phone
LDWS
ISA Advisory
ACC
Focus
Vehicle stability
Information to the
driver
Warning the driver
Aspects of longitudinal
and lateral control
Driver’s locus
of control
None
Full
Depends
Overrideable
System supplier
OEM
OEM aftermarket
Nomadic
OEM aftermarket
OEM
Safety issue
Technical
Distraction
Understandability
Controllability
Typical human
interface
None (or via existing
controls)
Screen + Audio
Buzzer, Symbol
Button
Small display
Existing controls
Fig.1. IVIS and ADAS function matrix
57
Human Centred Design for Intelligent Transport Systems
4 Integration of Methods
Humanist partners had several meetings and an internal workshop to discuss
the process of developing an integrated methodology. The main objective of the
workshop was to collate and review existing knowledge concerning the
integration of individual methods into a more holistic approach to assessing
safety and usability. A detailed report is in HUMANIST Deliverable E.3 [4]. An
Integrated methodology was defined by HUMANIST TF E as:
“Structured human factors evaluation (of a system or function) that
combines evidence from multiple assessments of different aspects of drivervehicle interaction within a conceptual framework”
Developing an integrated methodology is a complex task and TF E have made
some significant advances. From the first workshop it was clear that there
remains a research gap concerning how to combine individual methods into an
overall integrated methodology. The partners had a general common
understanding of safety and risk, but there was a need to identify and use a
common conceptual framework of driver information processing and risk. The
next challenge was to find a conceptual model.
An integrated methodology should help to:
1. Understand the system being studied
2. Identify behaviours and communication aspects that could have influence
on traffic safety
3. Formulate hypotheses concerning how (in what direction, in what aspects)
behaviour and communication might be influenced
4. Decide how the effects can be measured in principle
5. Design and undertake appropriate assessments & analysis
6. Combine and present results
There are several reasons why a conceptual framework is needed in context of
Integrated Methodologies. First, as a “theoretical backbone” it should offer some
“guidance” through the process of HMI assessment. Second, it should represent
relevant aspects of driver-vehicle interaction (with IVIS/ADAS), and third, it
should make meaningful predictions for user tests. From this point of view it was
obvious to clarify in a first step if some ideas could be derived from the work on
driver models performed by HUMANIST (Task Force C) at the “International
Workshop On Modelling Driver Behaviour In Automotive Environments”. The
model proposed by Oliver Carsten [5] was considered worth further
examination.
The group formed a view that the application of this model for the development
of an Integrated Methodology can be described as a top down approach. On the
other hand the question was raised in the discussion if such an approach would
be really viable in practice and it was suggested that a bottom up approach
which is driven by the actual demands of the system(s) under investigation
might be more appropriate. This is, of course, a strong point in particular
because the “Carsten-Model” is purely descriptive and does not specify how the
relationships described are moderated by IVIS and ADAS. As a result an
agreement on a conceptual model as specified in the definition of Integrated
58
Tools and methodologies for safety and usability
Methodologies could not be achieved within HUMANIST, however work
continued on integrated methodology.
The International Standards Organisation (ISO) have developed a suitability
standard. Additionally the RESPONSE [6] and ADVISORS [7] projects have
suggested methodologies which the HUMANIST group considered as a
framework for integration.
The aim of this part of the HUMANIST work was to merge all the methods
described in “The Matrix” towards one “Integrated Methodology” as a guide to
the assessment process.
A proposal of how an integrated methodology could be developed into a model
is shown below where available tools have been related to the different phases
of the process. The HUMANIST Group then applied their knowledge of what is
possible from current research, to establish where we are in the process of
developing an integrated methodology. The diagram below shows the model
that the HUMIST partners have generated.
As can be seen from the figure there are three phases in the process
(IVIS/ADAS description, selection of measurement methods, measurement and
analysis) where it is possible to refer to available inventories of methods as e.g.
the “HUMANIST Matrix”. The most critical item seems to be the Multi-CriteriaAnalysis or other combination/analysis tool at the end of the process where it
has to be concluded that suitable tools do not exist. Of course, some related
work has been done within the European projects ADVISORS and RESPONSE
but these were developed for different purposes. This does not mean to say that
an Integrated Methodology following this definition is incompatible with these
developments. Rather, the approach followed here aims at a stronger
microscopic or scientific level bridging the knowledge gaps between system
functions, driving behaviour and accident risk.
Task
Analysis
Describe
IVIS/ADAS
KEY
Available tools
Behaviour
Checklist
Identify potential
behaviours
Generate
Hypothesis
Stages of integrated
methodology process
Hypothesis
generation
tools
HUMANIST
Matrix
Select measurement
methods
Does not
exist
SUITABILITY
Framework
Measure,
Analyse, present
Some
Ideas
MultiCriteria
Analysis
Integrate results
Available
Fig.2. HUMANIST TFE model of the stages and tools required for an integrated
methodology
As can be further concluded from the figure above, there are two phases in the
process (identification of potential behaviours, generation of hypotheses) where
at least some ideas on necessary tools exist but the state-of-the-art is far from
59
Human Centred Design for Intelligent Transport Systems
satisfying. The discussions within TF E on conceptual models and frameworks
[3] open research issues and reflect the need for a specific, meaningful,
empirically validated but also parsimonious conceptual model. The continued
review of models of driving behaviour might provide further input. However, it
also became very clear during the discussions that these are too general to
have more than a heuristic value. This means that we might now be at a stage
where as well as practical tests assessments, deliberate theorising and careful
experimentation are necessary in order to convert the orange and red fields in
the Figure above into green ones.
A joint AIDE/HUMANIST workshop took place in March 2006. The Humanist
integrated methodology model was presented to AIDE and it received positive
approval.
Overall the work on Integrated Methodology has been useful at two levels. At
the more immediate practical level the work has provided support for
assessment by summarising the features of specific approaches. In addition the
conceptual work has allowed a dis-aggregation of the complex issue of
integrated methodology into a number of separate components that appear to
be amendable to further research.
5 Driver appropriation of its over time
TF E was required to investigate methodologies to evaluate the safety and
usability of ITS systems. One group of such methodologies includes methods to
assess driver appropriation of IVIS and ADAS systems. Therefore the next aim
of the HUMANIST work was to consider methods for the investigation of driver
appropriation processes over time. In this paper ‘appropriation’ is considered as
a specific case for integration into the general topic of methodologies (as
depicted in Figure 2). The entirety of behavioural changes in response to a
safety measure is generally referred to as ‘behavioural adaptation’ [8] [9]. An
Organisation for Economic Co-operation and Development (OECD) report
states that “Behavioural adaptations are those behaviours which may occur
following the introduction of changes to the road-vehicle-user system and which
were not intended by the initiators of the change. Behavioural adaptations occur
as road users respond to changes in the road transport system such that their
personal needs are achieved as a result. The adaptations create a continuum of
effects ranging from positive increase in safety to a decrease in safety” [10]. TF
E contributors agreed that appropriation does not only include observable
behavioural changes, but also changes in cognitive, regulatory and motivational
processes that underlie those observable behaviours. TF E contributors agreed
therefore to adopt a wider view regarding the nature of the adaptation
processes.
A workshop was held and several presentations from partners were given
providing different methods. A review of the workshop presentations and
discussions can be found in HUMANIST Deliverable E.5 [11]. In addition the
next deliverable, E.6, [12] provides a detailed description of 12 methods that
can be used to assess driver appropriation of ITS over time. Additionally brief
descriptions of a further 10 methods are provided with advantages,
disadvantages and comments on how the method is linked to driver
60
Tools and methodologies for safety and usability
appropriation. The TF has developed a main body of the deliverable that
provides an overview of driver appropriation processes and a discussion of
scope for an integrated methodology for appropriation measurement informed
by an underlying model of driver appropriation processes.
Starting from a number of presentations explaining possible methods to
measure driver appropriation, the TF built a better understanding of different
aspects of appropriation. This led to a deeper understanding and a better
definition of appropriation. Task Force E aimed to develop an ‘integration model’
to conceptualise integrated driver appropriation methods. The HUMANIST task
force agreed on the following definition of driver appropriation:
“Acquisition of knowledge, skills and attitudes underlying short term and
long term changes in behaviour”
As stated in this definition, driver appropriation in response to ADAS may occur
directly after introducing a new system into the vehicle (e.g. short-term changes
in workload), as well as after a certain period of time of driving with the system.
Furthermore, the definition emphasises the different nature of causes
underlying driver appropriation, which are knowledge (e.g. of how a system
works in different conditions), skills (e.g. in using /driving with the system), and
attitudes (e.g. trust in the system).
In order to conceptualise driver appropriation it was regarded as useful to view
different aspects of appropriation in relation to the three levels of the driving
task as proposed by hierarchical control models of driving (Michon, 1985).
Driver appropriation
Operational
Tactical
Strategic
ÆSkill
aquisition
Æ Use
Æ Knowledge and
attitudes
Æ Adaptation
of behaviour
Fig.3. A diagram to show the different levels of driver appropriation
Appropriation on the ‘operational level’ of the driving task is mainly concerned
with skill acquisition in how to operate a system. Decisions and skill in how to
use or to apply a system in particular situations and driving manoeuvres
(overtaking, lane changing) primarily affect driving on the ‘tactical level’.
Overall, behavioural adaptation in response to ITS is observable on the tactical
level of the driving task. Changes in drivers’ knowledge, attitudes and motives
(e.g. overall risk taking, trust in a system) will mostly affect the ‘strategic level’
of the driving task, and thereby influence control on the lower levels. As the
three hierarchical levels do interact with each other, adaptation at one level of
driver control has influence on the other levels as well.
A number of factors influence drivers’ response to ITS. HUMANIST TF E
focused on appropriation processes in response to IVIS and ADAS that are
61
Human Centred Design for Intelligent Transport Systems
influenced by three interacting factors: system characteristics, situation
characteristics and driver characteristics.
A conceptual model of driver appropriation has been developed (By Anke
Mogilka of CUT forming part of the PhD studies). The model together with a
detailed description can be found in HUMANIST Deliverable E.6 [10]. The
model aims to draw attention to relevant variables and their interdependences
in influencing driver appropriation. The model includes several processes,
including behavioural adaptation, risk compensation, and changes in
information processing in response to the introduction of ADAS.
According to this model, driver appropriation develops in response to changes
in the driving task due to driver assistance systems. The systems’
characteristics have a direct impact on driver’s information processing. The
degree of assistance offered by the system, i.e. the level of automation and the
systems feedback mediated by the HMI are believed to have an impact on
driver’s situational awareness and mental workload. Feedback from the system
about its current state and behaviour is necessary to build up a comprehensive
situation model that governs behaviour in unforeseen situations. Driver’s
workload is thought to have an influence on alertness and vigilance in such a
way that vigilance decreases to a greater extent over time when workload is
below or above a medium optimal level.
6 Integration of appropriation methods
Within the work of TF E, a concept was developed concerning possible
integration of methods based on the theoretical background and processes of
appropriation. In order to develop the concept/model the integrated
methodology model presented in a previous deliverable [3] was used as a
‘baseline’ approach. There were several issues raised and discussed. It should
be noted that previous work of TF E has been more concerned with methods
that relate to skill acquisition and driver behaviour measurement.
The new and wider perspective requires methods that have not hitherto
received as much attention including: elicitation of attitudes and knowledge by
“soft” means; extended field trials and naturalistic measurements; measurement
of system use/ non-use and measurement of changes in use of vehicles. A very
important issue is the need to be able to identify loss of ‘manual driving’ skills
caused by new technical system and the potential impacts of this skill loss e.g.
during system malfunction or failure.
The HUMANIST TF E proposed integration of driver appropriation model is
shown below. It is considered a working model that requires further research
and validation:
62
Tools and methodologies for safety and usability
Task
Analysis
Describe
IVIS/ADAS
Behaviour Checklist
Operational level
Tactical level
Identify skills
required
Identify potential
behaviours
Generate
Hypothesis
Qualitative and
communication
methods
Strategic level
Identify driver knowledge/
understanding and attitudes
Hypothesis generation
tools
HUMANIST
Matrix
Select measurement
methods
Measure,
Analyse, present
SUITABILITY
Framework
Integrate results
MultiCriteria
Analysis?
Fig.4. Diagram to summarise HUMANIST TFE proposed model of the
integration of driver appropriation methods
7 Conclusions and future work
HUMANIST TFE has made a significant research contribution through the
development of a matrix of methods and advances towards an integrated
methodology. The work on Integrated Methodology has been useful at two
levels. At the more immediate practical level the work has provided support for
assessment by summarising the features of specific approaches. In addition the
conceptual work has allowed a dis-aggregation of the complex issue of
integrated methodology into a number of separate components that appear to
be amendable to further research. It is therefore a challenge for future projects
to make progress in these target areas.
Discussion within TF E on driver appropriation has helped to understand the
requirement for future work in the research area. Starting from a “flat” list of
possible methods to measure driver appropriation, a better understanding of
different aspects of appropriation was built and a three-level characterisation
developed. This led to a deeper understanding and a better definition of
appropriation based on common understanding. Because driver appropriation
and the integration of appropriation methods is such a new work area more
questions have been raised than answers provided. The next phase of work in
this area should focus on further developing the conceptual model of driver
appropriation and the appropriation integrated methods model. In developing
these models further it will be important to resolve detailed issues. For example:
•
How can we measure transfer of learning from other environments?
•
How do differences in intelligence or educational background affect
appropriation?
63
Human Centred Design for Intelligent Transport Systems
•
What intervals should be measured to detect important adaptive changes?
•
How can we quantify driver distraction (internal and external) and its
impacts?
•
What are the driver appropriation processes for IVIS and ADAS and their
impact on safety?
The work of TF E has been very productive with some thought-provoking
contributions in this relatively new scientific area. These considerations led to a
better understanding of the development of innovative methodologies to
evaluate ITS safety and usability.
8 Acknowledgement
The Authors would like to thank all HUMANIST TF E participants for their
contributions. The authors are also grateful to the EC who provided funding for
the HUMANIST Network.
9 References
[1]
[2]
[3]
[4]
[5]
[6]
[7]
[8]
Gelau, C. (2004). Fahrerablenkung durch Informations- und
Kommunikationssysteme im Fahrzeug: Auswirkungen auf das
Fahrerverhalten und die Verkehrssicherheit. In B. Schlag (Hrsg.),
Verkehrspsychologie. Mobilität – Verkehrssicherheit – Fahrerassistenz (pp.
297-316). Pabst Science Publishers
European Commission (2004). HUMANIST project Deliverable D.2/E.2.
Impact of IVIS on driver workload and distraction: Review of assessment
methods and recent findings (pp. 1 – 104). All HUMANIST deliverables will
be located on the website (www.noehumanist.org)
European Commission (2007). HUMANIST project Deliverable E.4. Report
on the tools and methods for integrating methods into an assessment
methodology (pp. 1 – 50).
European Commission (2005). HUMANIST project Deliverable E.3. Issues
Related to Integrated Methodologies: Report of the Task Force E Santorini
Workshop, June 13, 2005 (pp.1 – 203).
Carsten (2005). From driver models to modelling the driver: What do we
really need to know about the driver? Proceeding of the International
Workshop on Modelling Driver Behaviour in Automotive Environments,
Ispra, Italy, May 2005
European Commission project RESPONSE. Response (2006) Draft of a
Code of Practice for the Design and Evaluation of ADAS, Contract: FP6507075 – Response, Deliverable 11.1
European Commission project ADVISORS. Advisors Final Technical
Report, Contract: 1999-RD.10047, Project: GRD1 2000 10047
Rudin-Brown, C. M., & Parker, H. A. (2004). Behavioural adaptation to
adaptive cruise control (ACC): Implications for preventive strategies.
Transportation Research Part F: Traffic Psychology and Behaviour, 7(2),
59-76.
64
Tools and methodologies for safety and usability
[9]
Sagberg, F., Fosser, S., & Sætermo, I.-A. F. (1997). An investigation of
behavioural adaptation to airbags and antilock brakes among taxi drivers.
Accident Analysis and Prevention, 29(3), 293-302.
[10] OECD (1990). Behavioural Adaptations to changes in the road transport
system. Organization for Economic Co-operation and Development: Paris.
[11] European Commission (2006). HUMANIST project Deliverable E.5.
Workshop on methodologies for the assessment of ITS in terms of driver
appropriation processes over time, December 12, 2006 (pp. 1 – 34).
[12] European Commission (2007). HUMANIST project Deliverable E.6.
Methodologies for the assessment of ITS in terms of driver appropriation
processes over time (pp. 1 – 75).
© Transport Research Laboratory 2007
65
Tools and methodologies for safety and usability
EVALUATING DRIVER MENTAL WORKLOAD
USING THE DRIVING ACTIVITY LOAD INDEX (DALI)
Annie Pauzié (INRETS, France)
annie.pauzie@inrets.fr
ABSTRACT: Methodologies are required to support the user-centred design
and evaluation of in-vehicle Information and Communication Technology (ICT).
This paper reports the outcomes of three on-road experiments conducted to
assess the usability of mobile phone and guidance/navigation systems. These
utilized a method for evaluating drivers’ mental workload: the Driving Activity
Load Index (DALI). For the guidance situation, functions were tested for two
systems corresponding to two levels of technology maturity. Outputs of the
DALI were used to identify which aspects of the old generation system had to
be improved, and how the new generation could be designed to be more usable
by drivers.
1 Introduction
Many research studies have been conducted to investigate the road safety
consequences of deploying Information and Communication Technology (ICT)
in vehicles [1],[2]. The objective is to create a balance between the potential
interference induced by these systems versus the potential benefits that can be
derived from them in supporting the driving task.
To evaluate these systems, it is necessary to have an efficient methodology that
is applied according to the type of function, the type of system and the context
in which the system is used [3]. A quite exhaustive overview of available
methodologies, tools and techniques has been conducted within the framework
of the network of Excellence HUMANIST [4]. Classically, the parameters that
have been taken into consideration for safety evaluation have been related to
the vehicle; for example, deviations from vehicle trajectories deriving from
system use [5]. Other parameters include drivers' visual strategies, visual
demand deriving from use of on-board screens [6], general driver behaviors
[7],[8] and overall driver workload according to the situation [9].
The assessment of workload is coupled with the difficulty of the task
experienced by the individual [10], because several reactions to task demands
are possible. The individual can adapt his behaviour to increased demand,
leading to a higher investment of effort but no perceptible effect on
performance; or, on the contrary, he can decide to change his strategy with a
lower level of performance. Then, moderate increases in task difficulty may
produce few observable changes in error rate, as the driver attempts to keep
performance constant by allocating more resources to the task [11].
Furthermore, inter-individual strategies are variable; some individuals develop
more effective strategies which require less effort to reach a level of
performance than do others. So, for all these reasons, objective performance
measures, which directly measure performance, are not sufficient by
themselves to evaluate the overall demands of a given situation.
67
Human Centred Design for Intelligent Transport Systems
Mental workload is a variable difficult to assess, in comparison with other
variables. Several methods have been developed to measure mental workload
[12]: measurements of physiological parameters, such as heart rate [13]; the
dual-task method [14]; and methods that elicit drivers’ subjective judgments
about the workload they have experienced. The latter include S.W.A.T
(Subjective Workload Assessment Technique; 15] and the NASA TLX - Task
Load Index [16]. Subjective measures are often used in practice because they
have many practical advantages over objective measures [17],[18],[19].
The next section of this paper discusses subjective methods for the evaluation
of drivers’ mental workload.
2 Subjective evaluation of drivers’ mental workload
2.1
Methods for subjective evaluation of mental workload
The subjective method allows for the evaluation rather than the measurement of
mental workload by comparing the perception of workload between situations. It
can therefore be regarded as a global, and even a "crude", criterion. Subjective
evaluation is often conducted in association with other workload measurement
techniques [20].
The SWAT is a sophisticated workload assessment tool, composed of a twostep process: in a scale development phase, data necessary to develop a
workload scale are obtained from individuals; during an event scoring phase,
people rate the workload associated with a particular task [21]. The primary
assumption of SWAT is that workload is function of three dimensions: time load,
mental effort load and psychological stress, each dimension having three
possible levels. All possible combinations of the three levels of each dimension
yield a 27-cell, three dimensional, matrix to represent workload.
The NASA TLX method assumes that workload is influenced by mental
demand, physical demand, temporal demand, performance, frustration level
and effort. After assessing the magnitude of each of these six factors on a
scale, the individual performs pair wise comparisons between these six factors,
in order to determine the higher source of workload factor for each pair. A
composite note quantifying the level of workload is set up by using both factor
rating and relative weights computed from the comparison phase.
The NASA-TLX has been tested and used by the army; being considered as
superior in terms of sensitivity than other methods and well accepted by the
operator [22].
The DALI (Driving Activity Load Index) is a revised version of the NASA-TLX,
adapted to the driving task [23]. As previously mentioned, mental workload is
multidimensional and, among other things, depends upon the type of loading
task. The NASA TLX was originally designed to assess pilot workload in the
aviation domain.
The basic principle of DALI is the same as that for the TLX. There is a scale
rating procedure for six pre-defined factors, followed by a weighting procedure
in order to combine the six individual scales into a global score. The main
difference lies in the choice of the main factors composing the workload score.
68
Tools and methodologies for safety and usability
For the NASA TLX, one of the factors to be rated is called the Physical Demand
component and is usually defined in the following terms : " How much physical
activity was required ? -pushing, pulling, turning, controlling, activating,...-" It
appears that this question would not be very relevant when considering the
driving activity where the control of the vehicle is quite automatic for an
experienced driver, and where maneuvers are not supposed to be physically
demanding in modern cars.
Another example is the Mental Demand component defined in the TLX as
follows " How much Mental and perceptual activity was required? - thinking,
deciding, calculating, remembering, looking, searching,...". This statement
covers both perceptual and cognitive aspects of workload, and we think it would
be interesting in the context of the driving task to be able to identify these
various modalities.
Finally, the evaluation of the Performance factor can be made using objective
data. The subjective rating of a good performance by the driver can show
discrepancies with the measured one, but this difference might be due to many
factors other than the mental workload itself - low or high self-esteem,
motivations to fit to the standard performance,...DALI was derived principally by asking various experts involved in driving task
studies to define which were, in their opinion, the main factors inducing mental
workload for people driving a vehicle equipped with an on-board system (car
phone, driving aid system, radio,...). This investigation led to the development of
six workload dimensions for DALI: Effort of attention, Visual demand, Auditory
demand, Temporal demand, Interference and Situational stress.
Key results of previous and recent studies using the DALI tool are summarized
in the following paragraph. An overview is provided of the advantages and the
limits of this task load index according to the purposes of the investigation and
the various contexts in which it is used.
2.2
Evaluation of driver’s workload using Hand-Free
mobile Phone
Using a mobile phone while driving raises the issue of road safety. Unlike other
Information and Communication technology developed to support the driving
task, the activity of phoning is disconnected from the driving task itself. There is
then no benefit of this function in terms of enhancement of the driving task
process for the driver. Use of the mobile phone while driving would induce a
high probability of interference in terms of attentional demand for the driver
[24],[25].
Nevertheless, some diversified findings are encountered in the literature, linked
to the modalities of experimental conditions. In order to test the load of
conversation, one author [26] used secondary verbal task techniques to assess
mental load while driving. According to this author, drivers showed no significant
changes in their driving performance while verbalizing, which made him assume
that the driving task priority was maintained as intended. Brown, Tickner &
Simmonds [27] used a verbal reasoning task based on grammatical
transformation in order to assess the effect of phoning while driving. Results
indicated increased errors in judgment of gaps, decreased skill in steering
69
Human Centred Design for Intelligent Transport Systems
through narrow gaps and decreased speed. Drory [28] measured driver
behaviour when using a mobile phone in a driving simulator. No serious
performance decrement was found concerning the driving activity, except when
the subjects actually dialed the number. Mikkonen & Backman [29] studied the
influence of phone conversation on driving performance in a familiar urban
environment. In this case, drivers paid more attention to their task, increasing
their alertness and their anticipation behavior. Tokunaga & col. [30] showed the
negative impact of the complexity of the phone conversation on reaction time
and on NASA-TLX values, both for young and older drivers.
In order to evaluate the efficiency of the DALI as a tool for the assessment of
mobile phone use, an experiment was carried out in a real road context (see
Pauzié & Pachiaudi [31] for detail of the experimental protocol and results). The
objective was to investigate evaluation of perceptual and cognitive load for the
driver in this phoning condition, knowing that specific factors of the DALI were
dedicated to these aspects. Results indicated that the global value of mental
load increased significantly when phoning and driving in comparison with the
reference situation corresponding to no system use. The load index was
significantly high for “auditory” and “interference” factors, in addition to “stress”.
The effort of “attention”, although higher than during a simple driving task, does
not increase significantly. So, in terms of subjective evaluation of the workload,
drivers identified the disturbance induced by phoning, through the perceptive
channel of audition, on managing the driving task, which induced stress.
Through this example, it is possible to illustrate how a tool based upon
subjective evaluation and driver awareness can allow for the understanding of
mental workload and cost of the task.
80
70
Not Phoning
Phoning
60
50
40
30
20
10
0
at t ent ion
visual
audit ory
t em poral
int erference
st ress
global
Fig.1. Factors and Global Value of the DALI for hand-free mobile phone use.
Advantages: Allow to better understand how the implementation of a system in
a vehicle can be experienced by driver (in this example, the most significant
factor was “interference” due to phoning while driving).
70
Tools and methodologies for safety and usability
Limits: Subjective evaluation gives information only on driver’s awareness about
workload he/her experienced in a defined context
2.3
Evaluation
of
driver’s
Navigation/Guidance function
workload
using
Navigation and guidance functions have been developed to support drivers at
the strategic level of the driving task, by supporting the navigation process.
They also support driving at the operational level, by supporting drivers to
anticipate upcoming maneuvers. Theoretically, driver reliance on auditory and
visual instructions to support their way finding decisions should decrease
mental workload and reduce driving errors. However, this objective can only be
reached if the system is correctly designed, that is to say avoiding
misconception, with correct timing for displayed messages, not too soon and
not too late, and clear, legible and visible visual information. Evaluation of
drivers’ mental workload, in addition to driving errors, is necessary to enhance
the effectiveness, usability and acceptability of systems developed to support
these functions [32],[33].
Two experiments conducted with a 9-year time lapse [34] are especially
revealing about the importance of making a distinction between the “benefit of
the function”, such as “instructions to guide the driver”, and “design of the
system for this function” such as “modalities of displayed instructions, timing of
auditory messages, legibility and understandability of messages”. The first
experiment described was testing the first generation GPS-based system; the
second was conducted using a new-generation version of the system, both
implemented by the same car manufacturer.
2.3.1 Drivers’ workload for old-generation guidance system
It was proposed that this system provide the option of auditory guidance
instructions or an electronic map display. The experiment was conducted in a
real road context and the purpose of it was to compare drivers’ workload for
these two options with a reference situation (that is, no system but a paper map;
Pauzié & Pachiaudi, 1997).
According to the DALI “global” score, use of the system corresponded to a
significantly higher workload for the driver in comparison with the “reference”
situation (guidance or electronic map). The DALI factors, “auditory” and
“temporal” demands were both critical factors, rather than “visual” load, with
high values for “stress” and “attention”. Based upon this data, it was possible to
determine that the messages displayed by this specific system, whatever the
option of guidance (display of arrows) or navigation (display of electronic map),
has a poor timing when delivering auditory instructions to the driver. These
results showed also that the navigation option induced a high rate of
interference, in comparison with the two other contexts of guidance and
reference situations.
71
Human Centred Design for Intelligent Transport Systems
80
70
Reference
Guidance
Navigat ion
60
50
40
30
20
10
0
at t ent ion
visual
audit ory
t em poral
int erference
st ress
global
Fig.2. Factors and Global Value of the DALI for Guidance and Navigation
Old generation system.
Advantages: Allow to identify the weakness of the design characteristics that
induced workload for the driver (in this example, poor timing of the auditory
messages)
Limits: Necessity to analyze objective variables such as driving errors to
complete the investigation (in this example, the value of driving errors could
modulate the conclusions regarding reference situation versus guidance
situation; it can be costly to use the system for the driver but it can induce less
driving errors).
2.3.2 Driver’s workload and new-generation of guidance system
Diversified contexts of orientation processes have been set up, varying
according to their level of workload for the driver, in order to test the validity of
the method DALI. Overall, 4 situations were identified, from HIGH to LOW
demand: “complex system” requiring cognitive and perceptive attentional
demand, “paper map” with no system, “guidance system” correctly designed
and “human co-pilot” giving instructions to the driver. The four situations have
been processed in real road contexts in a turnover order between drivers to
avoid effect of practice and learning [34].
According to the DALI global score, there is a significant difference between the
4 experimental sessions (Wilcoxon, Z= 3,007, p=0,003; Z= 2,224, p=0,026, Z=
2,539, p=0,011; Z= 3,923, p<0,001). More precisely, contrary to the previous
experiment, “use of guidance instructions” induced generally a lower workload
than “use of a paper map”, identified as “reference” in the previous paragraph.
Looking at the detail of the DALI factors, it appears that support of the system
for the driver is in terms of “stress”, “interference between driving and finding his
route”, “temporal”, “visual” and “attentional” demand, with significant differences.
Of course, ”auditory” demand was not rated by the driver in the context of the
paper map use.
72
Tools and methodologies for safety and usability
Hence, these results demonstrate that a guidance system correctly design in
terms of visual and auditory messages (timing, loudness, content) is an added
value for the driver by making the orientation task lighter in terms of cognitive
and perceptive processes. Furthermore, the DALI results showed that there is a
higher level of workload while using the system in comparison with relying on
the human co-pilot. Hypothesis can be made that this system could require a
phase of training longer than the timing of this experiment, in order for the driver
to be fully comfortable with the system. Additional testing with a longer training
phase could indicate if the system can be equivalent to a human co-pilot or not.
At least, the DALI results indicated that this system is superior to a paper map.
90
Com plex Syst em
Paper Map
Guidance syst em
Hum an co- pilot
80
70
60
50
40
30
20
10
0
at t ent ion
visual
audit ory
t act ile
t em poral
int erference
st ress
Fig.3. Factors and Global Value of the DALI for Guidance and Navigation.
New generation system
Advantages: Allow identifying correctly design functions of in-vehicle system,
able to support driving task, in comparison with situation with no system (in this
example, driving with a well- designed guidance system in terms of visual and
auditory messages (timing, loudness, content) induced less workload for the
following factors “attention”, “visual”,” temporal”, “interference”, “stress” in
comparison with the situation of paper map use).
Limits: Necessity to set up several contexts to be able to use this tool (reference
situation versus tested situation or several tested systems), as it allows relative
and not absolute results.
2.3.3 Driver’s workload and maturity of technology
The two previous experiments illustrated evolution of the design characteristics
of a system for a given function in relation to the improvement of the
technology. They highlighted the need to keep in mind, while testing in-vehicle
ICT, that both functions and system design can have an impact on usability and
safety, but at a different level. At this stage, and based upon literature
investigations, it can be considered that the guidance function by itself has a
positive impact on the driving task (less driving errors and hazardous behavior,
73
Human Centred Design for Intelligent Transport Systems
less workload in comparison with paper map use) as long as the guidance
instructions are correctly designed. Indeed, in this experiment, a specific care
on the HMI design of the guidance instructions was applied: enhanced contrast,
simplified visual display, well-paced auditory message. The objective was to
evaluate the added value brought by the guidance functions in an optimized
context of HMI design. The chosen HMI design was not the only solution, but
avoid poor interface such as monochrome screen (example: mobile phone on
the left of the figure 4) or complex map display.
To summarize, the use of the DALI can support the designer in identifying which
aspects of the driver-vehicle interface are inducing perceptual and cognitive
demand for the driver due to poor design of the system, and to improve next
generation of prototypes, allowing also to take into account the quick evolution
of the technology in this domain through the iterative processes of design and
evaluation (figure 4: examples of two successive generations of mobile phones
HMI).
Fig.4. Old and new generation of mobile phone giving guidance information to
the driver
3 Conclusion
The measurement of driver workload complements other workload metrics in
bringing additional information and allowing broader understanding about the
complex interactions between drivers and the systems they use while driving.
DALI, as a subjective workload evaluation tool, allowed for the gathering of data
that was usable by the designer in improving his system prototype. It allowed
enabled identification of the impact of a given system implementation by
comparing results with a reference situation with no system. One of the main
advantages of this tool is that it makes it possible to identify the origins of driver
workload, allowing for corrective action at the identified level (e.g., high
interference and visual load indicate that an in-vehicle system has a demanding
visual display). The possible design improvement would be to add factors linked
to specific aspects of the driving task useful to evaluate the impact of ADAS
functions (e.g. level of stress to keep distance with the vehicle ahead, in the
case of a system having an impact on this specificity of the driving task). It is
planned to conduct further investigations to improve this method by varying
these types of situations. The “DALI tool kit”, comprising the detailed method
and procedures for the automatic computation of statistics and the display of
74
Tools and methodologies for safety and usability
graphical outputs, will soon be available on the INRETS web site
(www.inrets.fr), allowing any researcher to use it in his/her scientific context.
4 References
[1]
[2]
[3]
[4]
[5]
[6]
[7]
[8]
[9]
[10]
[11]
[12]
[13]
[14]
Carsten, O.M.J. and Nilsson, L. (2001) Safety Assessment of Driver
Assistance Systems. European Journal of Transport and Infrastructure
Research, 1 (3). pp. 225-243. ISSN 1567-7141.
Vaa T., Gelau C., Penttinen M. Spyroupolou I., 2006, its and effects on
road traffic accidents - State of the art, 13 World Congress on ITS, London,
9 October 2006.
Bekiaris E. & Stevens A., 2005, Common risk assessment methodology for
advanced driver assistance systems, Transport Reviews, Volume 25,
Issue 3 May 2005 , pages 283 – 292.
Gelau C., Stevens A. & Cotter S., 2004, Impact of IVIS on driver workload
and distraction: Review of assessment methods and recent findings,
Deliverable D.2/E.2, HUMANIST NoE, 2004.
Zwhalen H.T.& Balasubramanian K.N. (1974). A theoretical and
experimental investigation of automobile path deviation when driver steers
with no visual input, Transportation Research Record, 520, 25-37.
Pauzié A., 2002, In-vehicle communication systems: the safety aspect,
Injury Prevention Journal; 8, 0–3.
Vitense H.S., Jacko J.A., Emery V.K., 2003, Multimodal feedback: an
assessment of performance and mental workload, Ergonomics, Volume
46, Numbers 1-3 / 15, Taylor & Francis, January 2003, pp 68 – 87.
Brookhuis K., de Waard D. and Janssen W., 2001, Behavioural impacts of
Advanced Driver Assistance Systems–an overview, EJTIR, 1, no. 3 (2001),
pp. 245 – 253.
Lansdown T.C., Brook-Carter N., Kersloot T., 2004, Distraction from
multiple in-vehicle secondary tasks: vehicle performance and mental
workload implications, Ergonomics, Volume 47, Number 1 / 15, Taylor &
Francis, January 2004, pp 91 - 104.
Gopher D. & Donchin E. (1986). Workload - an examination of the concept.
In K.R. Boff, L. Kaufman & J.P. Thomas ( eds.), Handbook of perception
and human performance. Volume II, Cognitive processes and
performance. New York : Wiley.
Zeitlin L. R., 1995, Estimates of Driver Mental Workload: A Long-Term
Field Trial of Two Subsidiary Tasks, Human Factors, Vol. 37, 1995
De Waard D. (1996). The measurement of drivers' mental workload, PhD
thesis, Traffic Research Centre VSC, University of Groningen, The
Netherlands, 125 p.
Casali J., and Wierwille W.A. (1984). "On the measurement of pilot
perceptual workload : a comparison of assessment techniques addressing
sensitivity and intrusion issues". Ergonomics, 27, 10, pp 1033-1050.
Shingledecker C. (1982). Performance evaluation of the embedded
secondary task technique. Aerospace Medical Association Annual
Scientific Meeting, 151-152.
75
Human Centred Design for Intelligent Transport Systems
[15] Reid G., Shingledecker C., Nygren T. & Eggemeier T. (1981).
Development of multidimensional subjective measures of workload,
International Conference on Cybernetics & Society, Atlanta, 403-406.
[16] Hart, L. A.,, Staveland, L., “Development of the NASA Task Load Index
(TLX): Results of empirical and theoretical research, In P. A Hancock and
N. Meshkati (Eds.), Human Mental Workload, pp. 139-183, Amsterdam,
North-Holland, 1988.
[17] O’Donnell R.D., Eggemeier F.T., 1986, New York: John Wiley and Sons,
Boff KR, Kaufman L, Thomas JP, eds. Handbook of Perception and
Human Performance. Volume II.
[18] Wierwille, W. W., & Eggemeier, F. T. [1993). Recommendations for
mental workload measurement in a test and evaluation environment.
Human Factors, 35, 263-281.
[19] Colle H.A., 1998, Context Effects in Subjective Mental Workload Ratings,
Human Factors, Vol. 40, 1998.
[20] Sheridan T. (1990). Human factors of driver-vehicle interaction in the IVHS
environment, Center of Transportation Studies, MIT, n° DTNH22-89-Z07595, NHTSA, U.S. Department of Transportation.
[21] Reid G. and Nygren T. (1988). The subjective workload assessment
technique : a scaling procedure for measuring mental workload". In Human
Mental Workload, P.A. Hancock & N. Meshaki (eds.), Elsevier, NorthHolland.
[22] Hill S., Lavecchia H., Byers J., Bittner A., Zaklad A. & Christ R. (1992).
Comparison of four subjective workload rating scales, Human Factors, 34,
4, 429-439.
[23] Pauzié A. & Pachiaudi G., 1997, Subjective evaluation of the mental
workload in the driving context, in " Traffic & Transport Psychology :
Theory and Application " , T. Rothengatter & E. Carbonell Vaya ( eds.), pp
173-182, Pergamon.
[24] Tokunaga R.A., Hagiwara T., Kagaya S. Onodera Y., 2000, Cellular
telephone conversation while driving : Effects on driver reaction time and
subjective mental workload, Human Performance : Driver Behavior, Road
Design, and Intelligent Transportation Systems. Annual Meeting of the
Transportation Research Board No79 , 2000, no 1724, pp. 1-6.
[25] Patten C, Kircher A., Östlund J. and Nilsson L., 2004, Using mobile
telephones: cognitive workload and attention resource allocation Accident
Analysis & Prevention, Volume 36, Issue 3, May 2004, pp 341-350.
[26] Wetherel A. (1981). The efficacy of some auditory-vocal subsidiary task as
measures of the mental load on male and female drivers, Ergonomics, 24
(3), 227-248.
[27] Brown I.D., Tickner A.H. & Simmonds D.C.V. (1969). Interference between
concurrent tasks of driving and telephone. Journal of Applied Psychology,
53, 419-424.
[28] Drory A. (1985). Effects of rest versus secondary task on simulated truck
driving task performance. Human Factors, 27 (2), 201-207.
[29] Mikkonen V. & Backman M. (1988). Use of the car telephone while driving.
Technical Report n° A39. Department of Psychology, University of
Helsinki.
76
Tools and methodologies for safety and usability
[30] Tokunaga R.A., Hagiwara T., Kagaya S. Onodera Y., 2000, Cellular
telephone conversation while driving : Effects on driver reaction time and
subjective mental workload, Human Performance : Driver Behavior, Road
Design, and Intelligent Transportation Systems. Annual Meeting of the
Transportation Research Board No79 , 2000, no 1724, pp. 1-6.
[31] Pauzié A. & Pachiaudi G., 1997, Subjective evaluation of the mental
workload in the driving context, in " Traffic & Transport Psychology :
Theory and Application " , T. Rothengatter & E. Carbonell Vaya ( eds.), pp
173-182, Pergamon.
[32] Ashby M.C., Fairclough S.H., Parkes A.M., A comparison of route
navigation and route guidance systems in an urban environment,
Proceedings of ISATA Conference, 1991.
[33] Pauzié A., Manzano J., 2007, Evaluation of driver mental workload facing
new in-vehicle information and communication technology, Enhanced
Safety in Vehicle Conference, Lyon, France.
[34] Pauzié A., Manzano J., 2006, Subjective evaluation of the driver’s
workload in real road experiment, AIDE Deliverable 2.2.6.
77
Tools and methodologies for safety and usability
COMPARISON OF SUBJECTIVE WORKLOAD
RATINGS AND PERFORMANCE MEASURES OF A
REFERENCE IVIS TASK
Tony Wynn (ESRI, United Kingdom),
John, H. Richardson (ESRI, United Kingdom)
T.wynn@lboro.ac.uk
ABSTRACT: The aim of this study was to establish subjective workload ratings
(NASA-TLX) for the ADAM surrogate reference task (SURT) and compare them
to ratings of a test battery of real IVIS tasks while participants performed a
surrogate of the driving task (Lane Change Task). The results indicated that
subjective workload ratings were comparable for both the real and reference
tasks, performance measures were significantly correlated with measures of
subjective workload and that performance of the secondary reference task
declined as task difficulty increased. Reasons for the pattern of results are
discussed.
1 Introduction
Traditionally car drivers have operated a number of ‘additional’ devices in the
driving environment including; radios, heating, ventilation and air conditioning
systems. Quite often these systems are controlled in addition to the primary
task of driving. Technological advances have facilitated the development more
sophisticated driver information systems intended to make travel more efficient
and less arduous [1]. The major concern with the introduction of new IVIS is that
they may introduce driver distraction sufficient to interrupt the primary driving
task [2].
Workload is a hypothetical construct [3] that represents the cost incurred by a
human operator in achieving a particular level of performance. It can be viewed
as the proportion of resources required to meet task demands. If the demands
of the task exceed the resources available to the skilled operator performance
will inevitably decline [4, 5]. Subjective workload ratings are the most common
method of obtaining estimates of the workload associated with a variety of tasks
and are an important consideration in the evaluation of performance. If
operators consider the workload of a task to be excessive they may behave as
though they are overloaded, even though the task demands are objectively low.
In such cases participants may report that the task was hard, but then perform
just as well as if undertaking a low workload task [6, 7].
A central theme in the development of IVIS is the evaluation of the user’s ability
to complete two or more tasks concurrently [8]. The primary aim of any
evaluation of IVIS is to ascertain whether drivers are able to perform the system
tasks in the dynamic and sometimes difficult circumstances that can be
expected during performance of the driving task. For these tests the primary
task does not necessarily have to be driving a car. It is only necessary to
present a primary task which demands an equivalent level of continuous
79
Human Centred Design for Intelligent Transport Systems
attention from the participant while operating a secondary task (the interface).
The dual task approach purposely divides the attention of the subject in order to
define the limits of their processing capabilities.
The rationale behind the use of reference tasks is that an IVIS device will
produce a particular pattern of behaviour and if a reference task produces the
same pattern of behaviour we can conclude that the reference is a suitable
surrogate. Reference tasks are intentionally generic in nature allowing the
impact of IVIS tasks to be assessed in a context independent fashion. A good
secondary reference task should be analogous to IVIS devices, complex
enough to illicit the same responses as a real IVIS yet, simple enough as to be
analogous to all IVIS. The ‘trade off’ in using reference tasks is between the
loss of ecological validity (real tasks), compared with the advantages of being
able to compare the results of studies more readily (standard tasks).
A robust secondary reference task was developed as part of the ADAM project.
The visual search task requires participants to report whether or not a prespecified target is embedded in a multi-item display. In this instance the
specified target is a circle that is distinguishable from other items (also circles)
within the display by size. Non-target items are designed to act as distracters.
Fig.1. Screen shot of the surrogate reference task (SURT).
Figure 1 illustrates how visual complexity of the task can be manipulated by
altering the size of the target, the thickness of the line and the number of
distracters present. The size of the target is always 44 arc minutes. In the left
image distracters are 22 arc minutes in which the target is easy to find, in the
centre image distracters are 35 arc minutes finding the target is moderately
difficult and in the right hand image the distracter size is 40 arc minutes, in this
instance the target is very difficult to distinguish. Task difficulty can therefore be
adjusted to match the difficulty (in terms of visual search) of a wide range of real
IVIS tasks.
The key aim of the current study was establish the subjective workload ratings
for an established reference task and a test battery of real tasks and then
consider these ratings with respect to the participant’s performance on the Lane
Change Task. This would then establish the utility of reference tasks as a
suitable surrogate for real tasks in the investigation of performance.
80
Tools and methodologies for safety and usability
2 Method
2.1
Participants
16 participants (2 Female, 14 Male) were selected at random from staff and
students at Loughborough University. Participants were required to have a full
United Kingdom driving licence and normal or corrected vision. They had an
average age of 35 years and 6 months and had held a driving license for an
average of 15 years and 7 months, self report estimate of annual mileage was
an average of 10,464miles.
2.2
Design
A within subjects ‘repeated measures’ design was used wherein each subject
completed each of the conditions. The trial design was counterbalanced so that
learning effects could be controlled for in the statistical analysis.
2.3
Lane Change Task
The Lane change Task is a laboratory control and event detection metric based
on the dual task paradigm. The dual task paradigm posits that primary task
performance will degrade with the introduction of a secondary task. In this case
LCT performance can be viewed as the primary task and it is designed to be
analogous to the driving task. It was developed as part of the ADAM project
(Advanced Driver Assistance Metrics [10].
Fig.2. Screen shot from the LCT. In this instance the driver has to change from
the centre lane to the right lane.
The LCT requires participants to negotiate a 3000m long section of three lane
highway. Participants are instructed by signs on the roadside (150m apart) to
perform a lane change manoeuvre (see figure 2). While completing the LCT
participants are required to perform a specific secondary task. To avoid speed
confounding the results it is controlled by the program and is kept at a constant
60kmph. The illumination reflects daytime driving with a constant light level.
Simulated low level engine noise is also provided. Visual information is
81
Human Centred Design for Intelligent Transport Systems
presented using an egocentric (front) view; no visual information is presented
regarding side or rear views.
The vehicle dynamics are such that the simulated car will behave as a standard
passenger car. Participants are required to change lane when instructed, when
not performing a lane change manoeuvre they are required to maintain a central
position within the lane. Performance of the lane change task by itself is used
as a measure of baseline performance for comparison with performance of the
LCT and a secondary task. During a LCT trial the LCT program automatically
records data to the computer on which it is running. From this data the LCT
analysis program can calculate a number of performance measures. These
include; mean deviation from the normative model, standard deviation from the
normative model, mean steering angle, as well as time course and distance
information to allow for standardisation of experimental runs. The normative
model is an ‘ideal’ vehicle path.
Fig.3. The LCT compares the normative model (solid line) to the participants
driving course (broken line)
Mean deviation from the normative model is the major variable of interest. It is a
measure of the effect of secondary-task demand and refers to the deviation
between the normative model and the actual driving course of the participant
along the track (see Figure 3 for a symbolic example of the normative model
and driving data). This deviation measure covers important aspects of the
driver’s performance; namely his/her perception (late perception of the sign or
missing a sign), quality of the manoeuvre (slow lane change results in larger
deviation) and lane keeping quality, which all result in an increased deviation.
2.4
In-Vehicle Information Systems
The IVIS used in this experiment was a Tom-Tom satellite navigation system
running under windows on a HP iPAQ (PDA). The ADAM SURT task was
presented on an 8 Inch TFT LCD Monitor produced by LinITX. Using this
equipment participants were required to complete seven LCT trials. Each of
these trials was dedicated to one of seven IVIS tasks (see Table 1) which have
been applied in previous research [10, 11].
82
Tools and methodologies for safety and usability
Table 1. List of Tasks
Task 1: PDA POI – entering a destination by selecting a “point of interest” using the PDA
Task 2: PDA address – entering a destination by “address” function using the PDA
Task 3: Shares short – searching for a share price from a single scrolling column of text
using an LCD screen
Task 4: Shares long – searching for a share price from three scrolling columns of text
using an LCD screen
1
Task 5: SURT circles task easy: - distracters are 22 arc minutes in which the target is
easy to find.
Task 6: SURT circles task average: - distracters are 35 arc minutes finding the target is
moderately difficult.
Task 7: SURT circles task difficult: - distracter size is 40 arc minutes; in this instance the
target is very difficult.
2.5
NASA Task Load Index (NASA-TLX)
The NASA-TLX [12] is a multidimensional tool for the measurement of
workload. It identifies a number of subjective factors that are relevant to
workload [13]. It has been extensively tested and widely used in the study of
human performance [14]. It provides an overall workload score based on a
weighted average of ratings on six subscales: mental demand, physical
demand, temporal demand, performance, effort, and frustration.
2.6
Procedure
Informed consent was sought from participants prior to commencement of the
experiment. Participants were informed of their right to withdraw at any time, for
any reason without penalty. Participants were invited to complete a practice
session (a block of 5 LCT trials lasting 15 minutes, instructions for the LCT were
provided) in order to familiarise themselves with the operation of the LCT. Upon
satisfactory performance of the practice task participants completed the
experimental trials. In the IVIS condition participants were required to complete
7 LCT trials, one for each task under dual task conditions. The order of these
trials was counterbalanced. The seven IVIS tasks are detailed in table 1. After
completion of each dual task participants were required to complete the NASATLX.
3 Results
Mean deviations from the normative model and participants NASA-TLX ratings
for the SURT tasks were significantly correlated (r (48) = .424, p< .01). There
was not a significant correlation between NASA-TLX ratings and tasks 1-4
(Real). A one-way repeated measures ANOVA was calculated for mean
deviation from the normative model on the LCT across the seven conditions
(PDA POI, PDA Address, Shares short, Shares long, SURT easy, SURT
1
The size of the target is always 44 arc minutes.
83
Human Centred Design for Intelligent Transport Systems
average and SURT difficult). The main effect was not significant [F (6, 15) =
1.029, P>0.05].
cu
iff i
ra g
SU
RT
-D
ve
lt
e
y
as
-A
RT
SU
PD
-E
o rt
sh
lo n
es
es
Sh
Sh
SU
RT
ar
ar
iff i
-D
ve
RT
SU
cu
ra
g
as
-A
RT
lt
e
y
I
O
-E
AP
SU
Ad
A
PD
PD
dr
es
s
or
t
g
Sh
lon
es
es
Sh
ar
ar
Sh
OI
0
g
0
RT
50
0.5
SU
100
1
s
150
AP
200
1.5
es
250
2
dr
300
PD
350
2.5
Ad
M ean N A S A -T L X S co re
400
A
M e a n d e v i a ti o n fr o m th e n o r m a ti v e
m odel
450
Condition
Condition
Fig.4. Average NASA-TLX rating and average mean deviation from the
normative model by condition
A repeated measures ANOVA was calculated for mean un-weighted NASA-TLX
score across the seven conditions (PDA POI, PDA Address, Shares short,
Shares long, SURT easy, SURT average and SURT difficult). The main effect
was significant [F (6, 15) = 14.421, P<0.05]. The subjective nature of workload
is illustrated in a significant test of between-subjects effects [F (1, 15) =
193.206, P<0.005].
3.1
Secondary task performance
50
12000
10000
40
MEAN TASK TIME
MEAN NO OF RESPONSE
45
35
30
25
20
15
10
8000
6000
4000
2000
5
0
0
ADAM Easy
ADAM
Average
CONDITION
ADAM Difficult
ADAM Easy
ADAM
Average
ADAM
Difficult
CONDITION
Fig.5. Mean number of responses and task time by SURT condition
A repeated measures ANOVA was calculated for mean number of responses in
the three SURT conditions (Easy, Average, and Hard). There was a significant
difference between the three conditions [F (2, 47) = 36.150, P< 0.005]. Similar
results were observed for total task time [F (2, 47) = 63.467, P< 0.005], mean
84
Tools and methodologies for safety and usability
reaction time [F (2, 47) = 13.388, P< 0.005], and errors [F (2, 47) = 14.302, P<
0.005]. The difference between number of key presses required to complete the
tasks and actual number of presses by participants was not significant [F (2, 47)
= 1.755, P> 0.05].
4 Discussion
Performance measures were significantly correlated with measures of
subjective workload as evidenced by the significant correlation between NASATLX rating and mean deviation from the normative model; this was the case for
the SURT tasks, but not the real tasks. Subjective workload ratings were
comparable for both the real and reference tasks. There were similarities in the
NASA-TLX profiles of the real and reference tasks. Figure 4 shows that the
subjective workload ratings obtained on the NASA-TLX for the reference tasks
were an approximate replication of the patterns observed in the battery of real
tasks. The ratings for the SURT easy task were similar to those for the shares
short task, the SURT average task rating was similar to ratings for the shares
long and PDA POI conditions and the SURT difficult rating reflected the rating
for the address entry condition. This suggests that the amount of workload
generated by the SURT tasks is reflective of the workload that would be
generated by a real task. The subjective nature of workload is illustrated by a
significant test of between-subjects effects. This result strongly suggests that
the ratings for each of the seven tasks were different for each participant.
These results reflected the hypothesised difficulty levels of these tasks at their
original conception. [11] evaluated the four real tasks (PDA POI, PDA Address,
Shares Short, Shares Long) used here using expert opinion and a key stroke
analysis. The panel of experts consisted of members of staff from TRL,
Nottingham University (UK) and the Chemnitz University of Technology
(Germany). The tasks were assessed on four criteria; input (how the driver
enters information), task (what needs to be done), display (what information is
presented), and output (what results are displayed by the system). The
negative, neutral and positive factors of each task were identified. All the tasks
were primarily visual however it was hypothesised that the Shares Short task
would be the easiest as it contained the least amount of visual information. The
Shares Long task contained only a small increment in visual demand (three
columns instead of one) therefore we would expect a similar increase in
workload rating. The PDA POI task contains an added element in that a
physical response is required therefore we would expect a further increase and
finally a further increment for the PDA address entry task which is more
demanding both visually and physically.
However, the performance measures do not support this. Scores for the mean
deviation from the normative model were not significantly different. We can
however examine the pattern of results with reference to [11]. What is surprising
are the scores for the mean deviation from the normative model for the real
tasks. In both the PDA and scrolling shares tasks the mean deviation was
higher than expected for the task rated as easier (PDA POI and Shares short) in
the expert review. We would have expected a smaller deviation from the
normative model in these conditions and therefore a significant difference.
85
Human Centred Design for Intelligent Transport Systems
However, there is a plausible explanation for this pattern of results. [15]
demonstrate task adaptation in which participants performance in a dual task
setting was shown to improve due to participants applying more effort to these
tasks and relaxing when performing the driving task alone. A similar effect may
be observed here in that participants may be mobilising more resources and
increasing effort in the more difficult conditions and/or relaxing and decreasing
effort while completing the easier tasks.
This possible explanation is particularly relevant in an examination of the SURT
tasks. As the difficulty of the SURT tasks increases so does the subjective
workload ratings of participants as illustrated by an increase in the mean NASATLX rating. In conjunction with this increase there is a decrease in the mean
deviation from the normative model. This is surprising given that this would be
indicative of good performance of the LCT. However, this point is clarified in an
analysis of the participants’ performance on the secondary task.
Data regarding performance of the secondary task suggests that as the difficulty
of the task increases participant’s performance declines. This is illustrated in
Figure 5 which shows that as the difficulty of the SURT task increases the
number of responses declines and the time taken to make these responses
increases. The results also imply that the SURT difficult task is too difficult for
participants and those participants disengage from the task; secondary task
performance is sacrificed to maintain what the participant perceives to be an
acceptable level of primary task performance. These results are in line with the
theory of compensatory effort [16] which states that when faced with a difficult
task participants will either mobilise more effort in order to achieve performance
goals (with an associated increase in physiological and behavioural costs) or
reduce their performance goals to avoid such costs. [6] imply that people are
aware of their task performance and that their estimates of subjective workload
are based on their perceptions of performance. It is reasonable to suggest that
in this instance participants are basing their estimates of workload on their
success at completing the SURT task and not on their performance of the LCT.
It is important to note that the SURT difficult task is still useful in performance
evaluation as it is representative of tasks that should probably not be
undertaken while driving.
5 Conclusion
The original aim of this study was to establish the subjective workload ratings
(NASA-TLX) for the SURT reference task and compare them to ratings of a test
battery of real IVIS tasks while participant’s perform a surrogate driving task
(LCT). The results of this study support the conclusions that subjective workload
ratings are comparable for both the real and reference tasks. Performance
measures suggest that participants trade-off secondary task performance in
order to maintain LCT performance at their perceived level of acceptable
performance.
86
Tools and methodologies for safety and usability
6 References
[1]
[2]
[3]
[4]
[5]
[6]
[7]
[8]
[9]
[10]
[11]
[12]
[13]
[14]
[15]
Beyreuther, G., and Laidebeur, A.: RTI, State of the Art Review, DRIVE
Project V1017 (BERTIE). Changes in Driver Behaviour Due to the
Introduction of RTI Systems, No. 28. 1989.
Wierwille, W. W.: Demands on Driver Resources Associated with
Introducing Advanced Technology into the Vehicle. Transportation Res (c),
1993, 1, (2), pp. 133-142
Gopher, D. and Donchin, E.: Workload- An examination of the concept, in
K. Boff, L. Kaufman and J. Thomas (eds.): Handbook of perception and
human performance (Wiley, New York, 1986).
Kahneman, D.: Attention and Effort. Englewood Cliffs, NJ: Prentice Hall.
1973
Wickens, C.D.: Multiple resources and performance prediction. Theor
Issues in Ergon Sci., 2002, 3(2), pp. 159-177.
Leggatt, A and Noyes, J.: Workload under complex task conditions in S. A.
Robertson (ed.): Contemp Ergon (1996; CRC Press).
Moray, N., Johanssen, J., Pew, R., Rasmussen, J., Sanders, A.F. and
Wickens, C.D.: Report of the experimental psychology group, in N. Moray
(ed.): Mental workload: its theory and measurement (Plenum, New York,
1979).
van Driel, C.J.G. Hoedemaeker, M. and van Arem B.: Impacts of a
Congestion Assistant on driving behaviour and acceptance using a driving
simulator, Transportation Res (F), 10, 2007, pp. 139–152
Bischoff, D.: Developing guidelines for managing driver workload and
distraction associated with telematic devices, 2007, NHTSA Paper No 070082
Mattes, S.: The lane-change-task as a tool for driver distraction evaluation.
In H. Strasser, K. Kluth, H. Rausch, and H. Bubb, (Eds.), Quality of work
and products in enterprises of the future. Ergonomia, Stuttgart, 2003, 57.
Horberry, T., Stevens, A., Robins, R., Cotter, S. and Burnett, G.:
Development of an occlusion protocol with design limits for assessing
driver visual demand, 2007; TRL, PPR 256
Hart, S.G. and Staveland, L.E.: Development of NASA-TLX (Task Load
Index): Results of empirical and theoretical research, 1988, In P.A.
Hancock & N. Meshkati, (eds.), Human Mental Workload, Amsterdam:
North Holland Press, pp. 239-250.
Moroney, W.F., Biers, D.W. and Eggemeier, F.T.: Some measurement and
methodological considerations in the application of subjective workload
and measurement techniques, Int. Jour. of Aviat. Psych., 1995, 5, 87–106.
Jorgensen, A.H., Garde, A.H., Laursen, B. and Jensen, B.R.: Applying the
concept of workload to IT work. In Virtual Proceedings of CybErg 1999:
The Second International Virtual Conference on Ergonomics, L. Straker
and C. Pollock (Eds.) (Perth, Australia: International Ergonomics
Association/Curtin University of Technology).
Matthews, G. and Sparkes, T.J. : The role of general attentional resources
in simulated driving performance. In Gale, A.G., Brown, I.D., Haslegrave,
87
Human Centred Design for Intelligent Transport Systems
C.M., Moorhead, I. & Taylor, S.P. (Eds.), Vision in Vehicles V. Elsevier
Science B.V, Amsterdam. 1996
[16] Hockey, G.R.J.: Compensatory control; in the regulation of human
performance under stress and high workload: A cognitive-energetical
framework, Biol. Psyc., 1997, 45, pp. 73-93.
88
Tools and methodologies for safety and usability
IMPAIRMENT OF LANE CHANGE PERFORMANCE
DUE TO DISTRACTION: EFFECT OF
EXPERIMENTAL CONTEXTS
Marie-Pierre Bruyas, Corinne Brusque, Adrien Auriault,
Hélène Tattegrain, Isabelle Aillerie, Max Duraz (INRETS, France)
marie-pierre.bruyas@inrets.fr
ABSTRACT: This paper aims to evaluate the consistency and sensitivity of the
Lane Change Test (LCT), which is subject of a proposed ISO standard. The
method aims at estimating driving demand while a secondary task is being
performed, by measuring performance degradation on a primary driving-like
task. An experiment was conducted in two experimental contexts, a driving
simulator and a Personal Computer (including pedals and steering wheel), and
with two auditory and two visual-manual secondary tasks. Three performance
measures were calculated: mean deviation adapted, correct lane change ratio
and lane change initiation. The effect of experimental context was significant.
The trajectory, measured by adapted mean deviation, was of better quality on
the simulator, while lane changes were initiated earlier on the PC. This
difference may be explained by the greater immersion of the driver in the driving
scene, which led to easier control of the trajectory in the simulator. Conversely,
participants initiated quicker responses to signs when using the PC, to the
detriment of trajectory control. The LCT was proven to be sensitive enough to
evaluate the driving performance impairment due to the simultaneous
performance of various secondary tasks.
1 Introduction
The handling and the use of In-Vehicle Information Systems (IVIS) while driving
induces dual-task situations that can lead to driver distraction. Distraction
occurs when a driver is delayed “in the recognition of the necessary information
to safely maintain the lateral and longitudinal control of the vehicle due to some
event, activity, object or person, within or outside the vehicle that compels or
tends to induce the driver’s shifting attention away from fundamental driving
tasks” [1].
Experimental studies point out differentiated impairment of driver performance
for auditory and visual-manual secondary tasks. Performing auditory tasks while
driving alters the ability to detect an event, to respond to it correctly and quickly
[2-5]. In the case of visual-manual tasks, the ability to control a vehicle is also
altered and lane keeping becomes more difficult [6-8].
Distraction has also been shown to increase the risk of having an accident. The
National Highway Traffic Safety Administration estimates that at least 25% of
police-reported crashes involve some form of driver inattention [9]. For Stutts et
al. [10], drivers were distracted at the moment of the crash in 8.3% of the policereports they analyzed. In-vehicle distraction is also reported as a contributory
factor in nearly 2% of the police fatal accident reports investigated by Stevens
and Minton [11]. However, these studies certainly underestimate IVIS
89
Human Centred Design for Intelligent Transport Systems
distraction as the reports are somewhat dated (1995-1999 [10], 1985-1995
[11]). More recently, in the US 100-car naturalistic driving study, Klauer et al.
[12] showed that drivers engaged in visually and/or manually complex
secondary tasks have a three -times higher near-crash and crash risk than
attentive drivers. Epidemiological studies confirm the negative impact of
distraction on road safety, showing that conversing by phone while driving (with
a hands-free or hand-held phone) is associated with a fourfold increase in crash
risk [13-14].
The risk of distraction generated by poorly designed device interfaces is thus of
main concern for authorities, designers and researchers. The challenge is to
develop simple and low-cost methods to assess the distraction of the HumanMachine Interface at the early stages of the IVIS design process.
The Lane Change Task (LCT) meets these requirements. This method was
developed for in-vehicle information systems evaluation in the German project
ADAM (Advanced Driver Attention Metrics) [15]. It aims at estimating driver
demand while a secondary task is being performed, by measuring performance
degradation on a primary driving-like task. This method is currently being
discussed in ISO working group TC22/SC13/WG8 as the basis of a standard to
access driver demand |16]. The LCT display can be implemented in different
experimental contexts (laboratory, driving simulator, mock-up or
complete/production vehicle) and should be applicable to all types of
interactions with in-vehicle information.
The aim of this paper is to evaluate the consistency and the sensitivity of the
LCT. Two main questions are addressed. Firstly, are the test results consistent
from one experimental context to another? Secondly, is the test sufficiently
sensitive to evaluate distraction effects of tasks with different characteristics?
To answer these questions, the LCT was implemented on two experimental
contexts: a driving simulator and a low-cost simulator (Personal Computer,
pedals and wheel) and with four very different secondary tasks: two auditory
tasks and two visual-manual tasks.
2 Method
2.1
Participants
Thirty participants were recruited and split into two equivalent groups of 15
drivers (7 males and 8 females in each group). All participants were aged
between 26 and 45 years (mean = 33.8; SD = 5.58). All of them had a driving
licence and reported driving at least 5,000 kilometres per year. All stated that
they drove several times each week.
All participants were used to using a mobile phone in everyday life (for more
than 7 years on average; SD = 2.58). Most of them also stated they used their
phone while driving, at least occasionally. All but one stated that they often
tuned the radio or dealt with a CD player while driving. None of them had
previous experience of the LCT.
90
Tools and methodologies for safety and usability
2.2
Procedure and design
Each group of 15 drivers performed the experiment in two sessions: one on a
driving simulator and the other on a Personal Computer (PC). One group began
with the PC condition (Order 1: PC 1st – Simulator 2nd) and the other with the
driving simulator condition (Order 2: Simulator 1st - PC 2nd). In between the two
sessions, there was a two week break.
On arrival, participants filled in a personal data questionnaire. They were then
given written instructions explaining the experiment. To begin each session,
participants performed a learning phase consisting of four runs without any
added task to enable them to become familiar with the LCT. The experiment
itself always began and ended with a run without an added task. Such a run
was also performed in the middle. These three runs provided the reference
data.
Four added tasks were also performed in four different orders to limit potential
task order effects. Each task was first executed without the LCT during about 90
seconds, to let participants get acquainted with it. Participants were then asked
to estimate its level of difficulty, by using a scale from 0 (no difficulty) to 10 (very
difficult). They were then invited to perform the task with the LCT. After a first
run, to ensure that the participants had understood the instructions, their
performance was registered. At the end, they had to estimate the difficulty level
of the task concurrently executed with LCT, using the same scale.
2.2.1 Driving context
For both sessions, the primary task was the LCT simulated driving task, which
involved driving along a straight 3-lane road at a constant speed of 60 km/h.
Participants were asked to use the steering wheel to maintain the vehicle
position in the centre of the indicated lane. Lane change signs appeared on
both sides of the simulated road to inform the drivers of the requirement to
change lanes. The signs were always visible but blank, until the lane change
instruction was given at a distance of 40 m before the sign position. Participants
performed 18 changes resulting in a duration of about 180 s per run (3,000 m).
Fig.1. The LCT track
91
Human Centred Design for Intelligent Transport Systems
The simulator session was conducted on the INRETS fixed base simulator in
Lyon, which had a front screen with a horizontal visual field of 50°. The car body
was a Renault Espace with a manual gearbox and all the standard passenger
compartment features, displays and controls. Software, conforming to the ISO
draft standard, was developed by the INRETS-MSIS team. Checks were carried
out in order to ensure a perfect correspondence between this software and the
LCT developed in the ADAM project [15].
The same software was used for the PC session, which was carried out on a
Personal Computer equipped with a Logitech “Gaming steering wheel and
pedals MOMO". The parameters of the transfer function between wheel and
vehicle were adjusted in order to make PC and simulator conditions
comparable. However the force feedback steering as well as the size of the
wheel remained different. The horizontal visual field was between 29° and 34°.
Such a configuration is very simple and not expensive making it possible to set
up an experiment very easily.
2.2.2 Secondary tasks
Two types of secondary tasks were performed: 2 auditory ones and 2 visual
manual ones.
Auditory tasks:
The first one (AT1) consisted of a series of statements pronounced by an
experimenter. The participants had to listen to each statement, to repeat it, and
to answer “Yes” if it was true, and “No” if it was false. For example: “Alice, my
mother in law is younger than her mother Martine” to which participants had to
respond “Yes”. Each assertion was randomly presented so that each driver had
a single series.
The second task (AT2) involved inventing sentences by creating a chain. To
begin with, the participants were given a first sentence comparing animals.
They then used the last word of the sentence to create the following one. For
example: “A horse is taller than a rat”, “a rat is smaller than an elephant”...
Visual manual tasks:
To perform the visual-manual tasks a screen was placed on the dashboard
(simulator condition) or fixed on the table (PC condition). Position and size
recommended by the ISO draft standard were respected for each display. A
numeric keypad was laid out under the screen so that the driver could carry out
the commands related to the two tasks (Figure 2a and 2b).
92
Tools and methodologies for safety and usability
a
b
Fig.2. LCT equipment set up in the simulator condition (a) and
the PC condition (b)
The first visual-manual task (VM1) was the Surrogate Reference Task (SuRT).
The objective was to look at a screen and to locate a circle among distractors
(smaller circles). To select the target, the participants moved a cursor to the
relevant zone by using right and left arrows. Then a button allowed them to
validate the choice and a new configuration was given by the system. Three
levels of difficulty could be activated. All drivers performed the “difficult” level
(Figure 3).
The second task (VM2) was the Critical Tracking Test (CTT) designed by
Dynamic Research (Figure 4). This task was of interest because the driver did
not control its evolution. Participants were faced with a moving black line. To
begin with, this line was displayed in the centre of the screen but then went up
and down. The objective was to keep it as close as possible to the centre part of
the screen, by using the up and down arrows of a keypad. Various difficulty
levels could be activated. All drivers performed an easy level (lambda = 1,
gain = 20).
Fig.3. Display for the SuRT with
difficult level
Fig.4. Display for CTT
93
Human Centred Design for Intelligent Transport Systems
2.3
Data analysis
2.3.1 Performance measurement
First, subjective evaluation of the task difficulty was assessed, by asking the
participants to estimate on a 0-10 scale the level of difficulty of each task, while
performed alone, and then while performed concurrently with the LCT.
Three measures of performance were also calculated:
•
The Mean Deviation Adapted: as specified in the LCT ISO draft, the
driving performance is mainly evaluated by the driver’s mean deviation
obtained for each run as compared with a normative trajectory. However,
the use of a unique model for all drivers has been discussed and a new
model is also being proposed [16-17]. The purpose is to define a reference
lane change trajectory for each driver based on his/her baseline runs. This
driver-tailored reference was used to calculate the mean deviation adapted.
•
The ratio of correct lane changes indicated the correctness of the
responses to lane change signs. Between two lane change signs an
observational zone was defined to determine the lane where the vehicle
was most frequently positioned. If this lane corresponded to the sign
indication, the lane change was considered as correct [16][18].
•
The lane change initiation corresponded to the distance between the lane
change sign and the distance at which the driver actually initiated it [19].
The calculation only applied to correct lane changes, as determined by the
method described above.
2.3.2 Statistical procedure
One-Way Repeated Measures Analyses of Variance ANOVAs were carried out
to analyze the data. The experimental design included one between-subjects
factor, the Session order (Order 1: PC1st_Simulator2nd; Order 2: Simulator1st
_PC2nd) and two within-subjects factors: the Experimental contexts (PC and
Simulator) and the Tasks (Reference, AT1, AT2, VM1 and VM2). Paired
comparisons were then computed with the Fischer LSD (Least Significant
Difference) test. For non-normally distributed data, non-parametric statistics
were used for significance testing. For both analyses, a significance threshold of
0.05 was accepted (p < 5%). The statistical procedures were performed with
SPSS.
3 Results
3.1
Subjective evaluation of the task difficulty
In terms of subjective evaluation of the task difficulty (Figure 5 1 ), no significant
difference was found between the two session orders (PC 1st–Simulator 2nd /
Simulator 1st-PC 2nd) - neither for the tasks being executed alone [F(1,
1
Whisker lines on figures 5 to 8 represent the standard deviation of each
variable.
94
Tools and methodologies for safety and usability
28) = 2.049, p = 0.163], nor for the tasks being executed with the LCT [F(1,
28) = 0.728, p = 0.401].
No experimental context effect was found; the subjective evaluation for a given
task did not significantly differ from the simulator condition to the PC condition,
when the tasks were executed alone [F(1, 28) = 0.041, p = 0.841] and when the
tasks were executed with LCT [F(1, 28) = 0.231, p = 0.635].
a
10
8
PC
Subjective evaluation
Subjective evaluation
10
Task performed alone
Simu
6
4
2
Task performed with LCT
8
6
4
PC
2
Simu
0
0
AT1
AT2
VM1
AT1
VM2
AT2
VM1
VM2
b
Fig.5. Subjective evaluation of the task difficulty executed alone (a) or with LCT
(b)
When executed alone, the task difficulty was not significantly different from one
task to another [F(3, 28) = 0.112, p = 0.953]. When performed concurrently with
LCT, the tasks appeared to have significantly different subjective levels of
difficulty [F(3, 28) = 28.612, p < 0.001]. Paired comparisons computed with the
Fischer LSD showed that all tasks differ significantly in terms of level of
subjective difficulty given by the participants [p < 0.005]. The tasks can then be
sorted, as follows, from the easiest to the most difficult: AT1, AT2, VM1 and
VM2.
3.2
Mean deviation adapted
For the measure mean deviation adapted (Figure 6), no significant difference
was found between the two session orders (PC 1st–Simulator 2nd / Simulator 1stPC 2nd) [F(1, 28) = 1.053, p = 0.314].
The analysis yielded a global experimental context effect; the mean deviation
obtained in the simulator condition was significantly lower than in the PC
condition. The trajectory was thus of better quality in the simulator condition
[F(1, 28) = 45.217, p < 0.001].
95
Human Centred Design for Intelligent Transport Systems
1.4
MeanDev Adapted
1.2
PC
Simu
1
0.8
0.6
0.4
0.2
Ref
AT1
AT2
VM1
VM2
Fig.6. Mean deviation adapted for each task and each experimental context
The mean adapted deviation was also significantly different according to the
tasks [F(4, 28) = 50.736, p < 0.001]. Paired comparisons computed with the
Fischer LSD showed that all tasks except AT2 and VM1 differed significantly in
terms of mean deviation [p < 0.005]. The mean deviation for the Reference
appeared to be significantly lower than for all other tasks and mean deviation for
VM2 was higher than for all other tasks. AT2 and VM1 were in an intermediate
position.
However, a significant interaction was registered between experimental
contexts and tasks [F(4, 112) = 5.489, p < 0.001], showing the effect of the
context on the distractive impact level of the tasks, especially on AT2 and VM1.
3.3
Correctness of the lane change
The ratio of correct lane changes was then computed for each task in each
experimental context (Figure 7). Since the ratios were not normally distributed,
non-parametric statistics were used for significance testing.
No significant difference was found between the two session orders (PC 1st–
Simulator 2nd / Simulator 1st-PC 2nd) [U(1, N=30) = 106.5, p = 0.806].
The analysis yielded a significant effect of experimental context, showing ratios
were higher in the simulator condition than in the PC condition [Z(1, N=30) = 2.738, p = 0.006].
A Friedman test revealed a significant main effect of the Tasks [χ2(4, N=30) =
72.082, p<.001]. Differences in terms of lane change correctness were found
between tasks except between Reference and AT1, between AT2 and VM1,
and between VM1 and VM2.
96
Tools and methodologies for safety and usability
Ratio of correct LC
1.2
1
0.8
PC
0.6
Simulator
0.4
Ref
AT1
AT2
VM1
VM2
Fig.7. Ratio of correct Lane Change for each task and each experimental
context
3.4
Lane change initiation
For the measure lane change initiation, no significant difference was found
between the two session orders (PC 1st–Simulator 2nd / Simulator 1st-PC 2nd)
[F(1, 28) = 2.289, p = 0.142].
The analysis yielded a global experimental context effect [F(1,28) = 12.670,
p = 0.001], showing participants took longer to initiate a lane change in the
simulator condition than in the PC condition.
Lane change initiation (m)
Ref
AT1
AT2
VM1
VM2
0
-5
-10
-15
-20
-25
PC
-30
Simulator
Fig.8. Lane Change initiation for each task and each experimental context
Lane change initiation was also significantly different according to the tasks
[F(4, 28) = 16.977, p < 0.001]. Paired comparisons computed with the Fischer
LSD showed that the reference gave significantly earlier LC initiation than all
other tasks while VM2 gave later LC initiation than all other tasks, except AT2.
No significant difference was found between AT1, AT2 and VM1.
97
Human Centred Design for Intelligent Transport Systems
4 Discussion
As the results did not show a significant effect of session order for the four
dependent variables, we shall focus on differences between experimental
contexts and tasks.
First, an effect of experimental context was obtained, showing differentiated
results according to the variables considered. Lane change performance was
better in the simulator than in the PC condition. More precisely, mean deviations
were smaller and the ratio of correct lane changes was higher. Santos et al. [20]
showed similar results by comparing secondary task impacts on driver’s lateral
control registered in a low cost laboratory driving simulator or a fixed-base
driving simulator. We can assume that, in the case of the driving simulator,
immersion of the participants in the driving scene is greater. The control of the
trajectory is thus easier for the drivers due to the increased visual realism. This
assumption is consistent with the findings of Kappé et al. [21] which show that
lane-keeping performance is improved by increasing the size of the horizontal
visual field and by the display of peripheral information. However, the quality of
the force feedback steering was also better in the simulator condition, which
may have contributed to the easier control of the trajectory registered in this
condition. A similar trend is highlighted by Mourant and Sadhu [22], who
showed that the feeling of immersion increases with the realism of force
feedback steering. Conversely, participants reacted earlier to the lane change
signs in the PC than in the simulator conditions, with distances of lane change
initiation being longer. In such a condition, participants seemed to pursue quick
sign detection and response, to the detriment of the quality of trajectory control.
It is possible that subjects behaved as if they were playing a video game in the
PC condition, due to the poorer realism of the driving task.
On the other hand, distractive effects of the four secondary tasks have been
evaluated with respect to their modalities: auditory or visual-manual and with
respect to their nature:
•
establishment of comparative judgement (AT1)
•
production of comparative sentence (AT2)
•
location of target among distractors (VM1)
•
realisation of a tracking test (VM2)
If executed alone, the four tasks were evaluated as being of a similarly low level
of difficulty; it was not the same in the dual task condition. AT1 was evaluated
as being the least distractive task, VM2 as the most distractive one, while AT2
and VM1 were evaluated at an intermediate level. Such a classification was
confirmed with the performance measures. The LCT method thus permits the
evaluation of the distractive impact of tasks of different characteristics. The
method is sensitive enough to evaluate driving performance impairment due to
auditory tasks even in the case of an easy one such as AT1. Moreover, if AT1
and AT2 were both concerned with comparative judgements and thus
presented some similarities, their reported difficulty and performance
impairments were not of the same level. This could be explained by the
necessity of producing complex sentences in the case of AT2. The LCT method
98
Tools and methodologies for safety and usability
also permitted the differentiation of distraction induced by two different visualmanual tasks. Both tasks differed with respect to participants’ involvement in the
control of their execution. In the case of VM1, the drivers controlled the pace of
the display. On the other hand, VM2 consisted of the stabilization of an unstable
component. Changes in the task were thus unforeseen and out of their control,
then required continuous reactions. These results highlight the joint effect of
interface modality and task nature on distraction.
Lastly, this study showed that the three driving performance measures were
affected by the experimental context even if the four tasks were organised into a
similar hierarchy. This result raises some questions about the possibility of
comparing distractive levels obtained in different experimental contexts. It is
thus necessary for each experiment to describe the equipment set up and to
use a set of reference tasks, in order to be better able to compare experiments.
5 References
[1]
Pettitt, M., Burnett, G., and Stevens, A.: 'Defining Driver Distraction', 12th
World Congress on ITS, San Francisco, US, November 2005
[2] Strayer, D. L., Drews, F. A., and Johnston, W. A.: 'Cell phone-induced
failures of visual attention during simulated driving', J. Exp. Psychol. Appl.,
2003, 9, (1), pp. 23-32
[3] Consiglio, W., Driscoll, P., Witte, M., and Berg, W. P.: 'Effect of cellular
telephone conversations and other potential interference on reaction time
in a braking response', Accident Anal. Prev., 2003, 35, (4), pp. 495-500
[4] Patten, C. J. D., Kircher, A., Ostlund, J., and Nilsson, L.: 'Using mobile
telephones: cognitive workload and attention resource allocation', Accident
Anal. Prev., 2004, 36, (3), pp. 341-350
[5] Bruyas, M.-P., Chapon, A., Lelekov-Boissard, T., Letisserand, D., Duraz,
M., and Aillerie, I.: 'Évaluation de l’impact de communications vocales sur
la conduite automobile (An evaluation of the impact of vocal
communication on the driving task)', Recherche Transports Sécurité, 2006,
91, pp. 99-119
[6] Engström, J., Johansson, E., and Ostlund, J.: 'Effects of visual and
cognitive load in real and simulated motorway driving', Transport Res. Part
F, 2005, 8, (2), pp. 97-120
[7] Ranney, T. A., Harbluk, J. L., and Noy, Y. I.: 'Effects of voice technology on
test track driving performance: implications for driver distraction', Hum.
Factors, 2005, 47, (2), pp. 439-454
[8] Tornros, J. E. B., and Bolling, A. K.: 'Mobile phone use--Effects of
handheld and handsfree phones on driving performance', Accident Anal.
Prev., 2005, 37, (5), pp. 902-909
[9] Ranney, T. A., Garrott, R., and Goodman, M. J.: 'NHTSA driver distraction
research: past, resent, and future', 17th ESV Conference, Washington DC,
2001
[10] Stutts, J. C., Reinfurt, D. W., Staplin, L., and Rodgman, E. A.: 'The role of
driver distraction in traffic crashes' (Chapel Hill Highway Safety Research
Center, 2001)
99
Human Centred Design for Intelligent Transport Systems
[11] Stevens, A., and Minton, R.: 'In-vehicle distraction and fatal accidents in
England and Wales', Accident Anal. Prev., 2001, 33, (4), pp. 539-545
[12] Klauer, S. G., Dingus, T. A., Neale, V. L., Sudweeks, J. D., and Ramsey,
D. J.: 'The Impact of Driver Inattention On Near-Crash/Crash Risk: An
Analysis Using the 100-Car Naturalistic Driving Study Data' (Virginia Tech
Transportation Institute, 2006)
[13] Redelmeier, D. A., and Tibshirani, R. J.: 'Association between cellulartelephone calls and motor vehicle collisions', N. Engl. J. Med., 1997, 336,
(7), pp. 453-458
[14] McEvoy, S. P., Stevenson, M. R., McCartt, A. T., Woodward, M., Haworth,
C., Palamara, P., and Cercarelli, R.: 'Role of mobile phones in motor
vehicle crashes resulting in hospital attendance: a case-crossover study',
BMJ, 2005, bmj.38537.397512.55
[15] Mattes, S.: 'The lane change task as a tool for driver distraction
evaluation', in Strasser, H., Rausch, H., and Bubb, H. (Eds): 'Quality of
work and products in enterprises of the future' (Ergonomia Verlag, 2003)
[16] ISO/TC22/SC13/WG8: 'Road vehicles - Ergonomic aspects of transport
information and control systems - Simulated lane change test to assess
driver demand', working draft, 2007
[17] Tattegrain-Veste, H., and Bruyas, M. P.: 'Adaptation of LCT measurement
using an Adapted curve for each subject', Driver Performance Metrics
Workshop, Ottawa, Canada, October 2006
[18] Engström, J., and Markkula, G.: 'Effects of Visual and Cognitive Distraction
on Lane Change Test Performance', Driving Assessment, Stevenson,
Washington, July 2007
[19] Harbluk, J. L., Burns, P. C., Lochner, M., and Trbovich, P. L.: 'Using the
Lane-Change Test (LCT) to Assess Distraction: Tests of Visual-Manual &
Speech-Based Operation of Navigation System Interfaces', Driving
Assessment, Stevenson, Washington, July 2007
[20] Santos, J., Merat, N., Mouta, S., Brookhuis, K., and de Waard, D.: 'The
interaction between driving and in-vehicle information systems:
Comparison of results from laboratory, simulator and real-world studies',
Transport Res. Part F, 2005, 8, (2), pp. 135-146
[21] Kappé, B., van Erp, J., and Korteling, J. E.: 'Effects of Head-Slaved and
Peripheral Displays on Lane-Keeping Performance and Spatial
Orientation', Human Factors: The Journal of the Human Factors and
Ergonomics Society, 1999, 41, pp. 453-466
[22] Mourant, R. R., and Sadhu, P.: ‘Evaluation of force feedback steering in a
fixed based driving simulator’, 46th Annual Meeting of Human Factors and
Ergonomics Society, Baltimore, Maryland, USA, October 2002
100
Tools and methodologies for safety and usability
CONNECTING BLACK BOX DATA AND DRIVING
BEHAVIOUR OBSERVATION FOR BETTER
UNDERSTANDING OF DRIVING BEHAVIOUR
Clemens Kaufmann, Christine Turetschek (FACTUM OHG, Austria)
clemens.kaufmann@factum.at
ABSTRACT: This paper presents two studies where the observation method
"Viennese Driving Test" (Wiener Fahrprobe) was connected to the data of a
Black Box (BB). The "Wiener Fahrprobe" was used to decode these BB data in
order to identify special driving behaviour like hard braking, fast acceleration,
sharp turning, etc.. Data from observation rides with novice drivers and self
reporting test rides on standardised test routes in Brno and Vienna were
analysed in this respect. With the help of the results gathered from the
evaluation recommendations for future research and future application were
established.
1 Introduction
Within two projects “Black Box – Novice Drivers” and “Black Box –
Introspection” the observation method “Wiener Fahrprobe” was connected to
the data of a Black Box (BB). Nowadays BB are used mainly for crash
investigations but not for studying driving behaviour, although it might be a quite
easy and cheap way to collect information about the driving style of individuals.
The Wiener Fahrprobe should help to decode these BB data in order to define
special driving behaviour like hard braking, fast acceleration, sharp turning, etc..
The future goal is to analyse driving behaviour by collecting BB data,
understanding those data, and additionally give feedback to the driver or maybe
the driving instructor based on the data, to make suggestions about
improvements or information and/or training needs. The advantages could be
that novice drivers could get a quick check on their driving behaviour by looking
at the data of the BB. But not only for novice drivers improvements could be
seen. Also while training or testing ITS applications the concrete interpretation
of driving behaviour will help to assess the effects on the driving behaviour of
such applications.
Several aims were defined at the beginning of the project:
1. Connecting both methods – BB & Wiener Fahrprobe
2. Find out what driving aspects BB data do reflect compared to data of the
driving behaviour observation done by human observers
3. Identify special driving behaviour with the help of BB data (for instance fast
acceleration, hard braking, fast turning)
4. Give recommendations for future research
101
Human Centred Design for Intelligent Transport Systems
5. Future goal: Improve driver education based on BB data that are easy to
collect, given that thorough interpretation is possible
6. Future goal: Evaluate differences in the driving style with or without using
ITS
To find an answer to all these questions a basic BB was used within the cars
which were used for the observation of novice drives and the introspection
rides. The BB registered accelerations in three dimensions. In parallel, either
the driver should be observed (novice drivers) or he/she should observe
himself/herself and note whether he/she remembered significant situations
(introspection).
This paper is based on an internal HUMANIST paper [1].
2 Methodology
2.1
Viennese Driving Test (Wiener Fahrprobe)
The Viennese Driving Test [2, 3, 4] is an instrument for behaviour observation of
drivers. This method gives researchers the possibility to get a structured
impression of the driving behaviour of a person. The observation was done by
one observer inside the car, sitting in the back of the vehicle. The observed
person was asked to drive along a standardised route of between 17 to 20 km
in length which included densely inhabited areas, and to a lesser degree rural
roads and motorways. The test route was divided into sections such as
intersections, motorway entrances or exits, sections of road between
intersections, roundabouts, etc. in order to make the evaluation of the behaviour
observation easier.
The behaviour variables were observed both in a standardised and in a nonstandardised way: The observer used an observation sheet where erroneous
types of behaviour were listed. One sheet per section had to be filled in. But
also non-standardised variables like errors, explicit interaction/communication
processes and traffic conflicts were noted. For each event a note had to be
taken and the number of the sections where the event happened was added.
2.2
Black Box (three axes acceleration recorder)
Acceleration is one of the most accurate indicators for critical driver behavior.
Based upon this awareness, CDV (Centrum dopravního výzkumu – Brno,
Czech Republic) developed a very basic Black Box (BB) which saves data of
acceleration on three axes {x, y and z} in relation to the time line.
The heart of the recording device (Black Box) is the Micro Electro Mechanical
System (MEMS sensor) that works as a tri-axes digital output linear
accelerometer. It includes a sensing element, and an IC interface element able
to take the information from the sensing element and to provide the measured
acceleration signals to the external world through an I2C/SPI serial interface.
The sensing element capable of detecting acceleration is manufactured using a
dedicated process called Thick Epi-Poly Layer for Microactuators and
Accelerometers (THELMA) to produce inertial sensors and actuators in silicon.
102
Tools and methodologies for safety and usability
The Integrated Circuit (IC) interface is instead manufactured using a
Complementary Metal Oxide Semiconductor (CMOS) process that allows a high
level of integration to design, and a dedicated circuit which is factory trimmed to
better match the sensing element characteristics.
The MEMS has a full user selectable scale of 2g, 6g and it is capable of
measuring accelerations over a maximum bandwidth of 2.0 KHz for the X, Y
and Z axis. The device bandwidth may be selected accordingly to the
application requirements.
The BB was located below the driver seat. The data were transferred and
logged on a Laptop after the rides.
There were different reasons for using "only" a basic BB within the project:
•
The project was designed as a pilot study to see if the connection between
the two data gathering methods is worthwhile
•
The costs of project has to be as low as possible
•
The BB has to be easily transferred to different cars which were used in
Vienna and Brno
2.3
Procedure of both projects
Both methods which are described here were connected to two projects.
Project Black Box – Novice Drivers:
This project was part of a dissertation carried out in the framework of the NoE
HUMANIST [5]. In the frame of the dissertation several novice drivers were
observed with the method of the Wiener Fahrprobe while driving a BB equipped
car on four different test routes in Brno, Czech Republic and in Vienna, Austria.
During these test rides data were registered with help of the BB.
The sample included 24 novice drivers in Brno (43 rides) and 7 novice drivers in
Vienna (14 rides). All test rides were done during August and September 2006.
Project Black Box – Introspection:
Additional to the test rides with novice drivers, introspection rides were done in
Brno as well as in Vienna. Employees of CDV and FACTUM drove on two of the
test routes with a car which was equipped with the BB. The drivers were also
observed and after each test ride the drivers recorded all noticeable situations
themselves. The aim was, on the one hand, to compare the observations of the
observers with the notices of the drivers and, on the other hand, to compare
both notices with the data from the BB. This procedure should help to get more
knowledge of the character of the BB data and what they mean in terms of
actual behaviour in traffic.
Two test rides were carried out in Brno in October 2006 and January 2007 and
two test rides were carried out in Vienna in February 2007.
103
Human Centred Design for Intelligent Transport Systems
The results and recommendations derived from both projects are reported in the
following chapters.
3 Results and achievements
According to the aims defined at the beginning of the project, the following
results can be reported so far.
3.1
Connecting both methods
To start with, it can be stated, that both methods could easily be connected. The
BB only has to be implemented in the car before starting the observation rides.
There is no further need for handling the BB since the data are recorded
automatically, then. The use of the BB while observing driving behaviour is
totally unobtrusive, it does not interfere with the behaviour of the driver, nor the
one of the observer. Just the time of the beginning and the time of the entrance
in a new section have to be noted by the observer on the observation sheet.
This extra effort is necessary in order to make the connection of the BB data
with the behaviour observation data possible.
During the project it turned out, that many "manually" registered behaviour
aspects could also be identified with the help of the BB data. Types of
behaviour recorded with both methods reflect acceleration, braking and turning
performances. Additionally, with the BB data the observer receives a simplified
impression of the overall driving style. As the observer never notices
acceleration, braking and turning behaviour in a such a continuous and
differentiated way as the BB can do, the BB reflect more thoroughly whether a
person drives smoothly or more in a staccato style. Because of this extra
information a more quantitative assessment of the driver is possible.
On the other hand, it is not possible to interpret the BB data without any
additional information. E.g., the amplitudes in the graphs (see in next chapter)
do not tell whether a person just drove into a curve very fast, or if he/she
approached an unexpected object and had to swerve rather quickly.
Thus braking and turning performances that can be observed (Wiener
Fahrprobe), can also be identified in the BB data, but BB data do not give any
information about circumstances.
3.2
Driving behaviour reflected by the BB data
In relation to what was said above, the first step was to find out the meaning of
the BB data, i.e., in which way do the data reflect observable driving behaviour.
Graphs of about 15 test drives were produced which display the acceleration in
the longitudinal axis (x-values) and lateral acceleration (y-values). The reason
for using the data from only 15 test rides was that within the financial and
schedule frame of this pilot study only a small sample of the test rides could be
evaluated. Furthermore we did not deal with z-values (vertical accelerations,
e.g., when driving over a hump) at this stage because of the same reason
stated above. The data used for evaluation within the novice driver project was
gathered from test rides with only one car.
104
Tools and methodologies for safety and usability
If the amplitude on the longitudinal axis of the values goes up this reflects
acceleration, and if goes down this means braking. Of course, also stand-stills
can be identified, for instance because of a red traffic light.
With the help of those graphs the first statements about the driving behaviour
can be made, for instance whether the acceleration behaviour of a driver is
steady or not. If the amplitude goes rather steeply up and down all the time this
means that a rather strong acceleration is always followed by a rather sharp
braking manoeuvre and vice versa. Behaviour observation can give the
additional information that, for instance, a hard braking of the driver occurred
due to a red traffic light.
If the amplitude on the lateral axis of the values goes up it shows a left turn of
the car, if it goes down it reflects a right turn.
The combined graph below shows the x- and y- values and displays the driving
behaviour with respect to acceleration and turning.
Two situations could be described with the help of this chart. The black circle
shows the acceleration and braking behaviour before and after two right turns.
First, the driver slows down before the right turn (seconds 33 till 35 – black line)
Æ turns right (seconds 35 till 37 – white line) Æ accelerates after the turning
manoeuvre (seconds 37 till 42) Æ brakes again for the next turn (seconds 43 till
46) Æ turns again right and accelerates during the turn (seconds 50 till 55).
The second situation describes the behaviour of the driver after a red traffic
light, marked with the white circle. The driver waits at the traffic light for about
20 seconds (seconds 112 till 132) Æ after the traffic light has turned green the
driver starts to accelerate and drives into the intersection (seconds 133 till 136 –
blue line) Æ slows down as a preparation for the turn, turns left and accelerates
during the turn (seconds 136 till 140).
Graph 1: Acceleration to the front (x-values) and acceleration to the side
(y-values) for 150 seconds
By looking only at the graphs it is of course not clear where on the test route
these situations occur. Therefore several self-introspection rides were carried
out with employees of the two institutions working within the project. The aim of
these rides was to describe the test routes according to turns, intersections,
105
Human Centred Design for Intelligent Transport Systems
etc., and the driven time between those sites. Also some hard braking and fast
acceleration manoeuvres were carried out on purpose in order to identify how
such behaviour is displayed by the BB data. This is necessary in order to be
able to interpret the graphs, because it makes clear what the amplitudes mean
in reality. With the help of these descriptions the sites that test persons have to
drive through can be identified. In this way special events like hard braking or
fast acceleration because of, e.g., an error of the driver become better
identifiable, by amplitudes on the x- or the y- axes in sections where there
should not be any deflections.
In a next step the charts we had received were analysed accordingly. The
following graph 3 shows 150 seconds of the test route. Graph 2 below displays
acceleration and braking behaviour, and right or left turns. The black circles in
both graphs show accelerations and brakings for "special" occasions, while the
grey circles show the turning behaviour.
By comparing the graphs derived from the BB data (graph 3), the notes of the
driving behaviour observation (table 1), and the map showing the test route
(graph 2), turning behaviour at intersections, accelerations and brakings before
and after these intersections, and also standstills at traffic lights, can be
identified.
Graph 2 & 3: Map of the first part of the test route in Vienna, acceleration in the
longitudinal axis (x-values) and lateral acceleration (y-values) for 150 seconds
Source: www.wien.gv.at [6]
106
Tools and methodologies for safety and usability
3.3
1
First right turn
1
Acceleration after first right turn
2
Second right turn
2
Braking before second right turn
3
Third right turn
3
Acceleration after second right turn
4
First left turn
4
Acceleration after third right turn
5
Acceleration during first left turn
Identification of "special" driving behaviour within the
BB data
As said, not only the behaviour before and after turns can be identified, also
special situations can be identified when comparing BB data and behaviour
observation. For a demonstration, a hard braking manoeuvre was selected. The
following graph shows a cut-out of the data from a novice driver and describes a
hard braking manoeuvre before a traffic light (black circle). The observer
marked this behaviour in her notes as "maladjusted speed before traffic light".
Several other situations like the above described were found in the BB data as
well as in the observation notes.
Graph 4: Acceleration on the longitudinal axis (x-values) and lateral acceleration
(y-values) for 300 seconds
107
Human Centred Design for Intelligent Transport Systems
3.4
Problems
Unfortunately, not all graphs can be interpreted that easily, and the comparison
with the behaviour observation notes and the street map did not work for all
drivers. Also the comparison of different charts of drivers displaying the same
section of the route was hard to interpret. The main problem was that the BB
used at this moment is only saving acceleration data on three axes and the
corresponding time (seconds), but cannot take up and reflect the exact position
of the driver. Therefore, from the data it is sometimes not clear where exactly on
the route the driver is located, and what the circumstances there are. Thus, the
comparison with other drivers becomes complicated because test persons
drove with different speeds and therefore arrived at sites at different times. Also
the fact that the test persons have to wait at a different number of traffic lights
makes comparisons of the graphs difficult. For example, some drivers were at
the first left turn within two minutes, while others needed two and a half minutes
to get there. The longer the test persons had been driving, the larger the bias
became. Without information about the driven meters, or GPS data, this
problem cannot be solved.
4 Recommendations for future research
Because of the results of these two projects and with the knowledge gained
about how to interpret the BB data with the help of the Wiener Fahrprobe, it
became clear that further research is needed. Especially, the preconditions for
comparison of the data of different test rides should be improved in the future.
The reason for problems there was that no continuous variable which would
make the data comparable, such as meters or other positioning parameters,
could be included in the BB measurements. Furthermore, there are still some
more questions about the correct meaning of the data to be answered. For
example: What does the difference between an amplitude which goes up to the
value of 2g and an amplitude which goes up to the value of 3g mean? Would
only a value of 5g mean that the behaviour was "really" inappropriate? Where
does the boarder go?
Because of these questions several recommendations for further research were
formulated. The aim of these recommendations is to make the data collection
more precise and better comparable in order to be able to interpret the outcome
of the data more thoroughly:
Recommendations concerning Data collection
The knowledge of the actual speed will help to interpret the data. Speed is one
criterion for the classification into appropriate and inappropriate behaviour.
Similar to speed, also a distance measurement will be needed. As already
said comparisons of the graphs were sometimes difficult because some of the
test persons drove faster or were waiting at red traffic lights more often than
other test drivers. With the knowledge of the driven meters the data of all test
rides could be compared more thoroughly and easily. The most efficient, but
also the most costly solution would be to include GPS and/or XL-meter data in
the data collection of the BB.
108
Tools and methodologies for safety and usability
•
During the analyses it also become obvious that the measuring of data was
not well calibrated, also depending on where the BB was put into the car or
which car was used. Especially a comparison between the data from the
novice driver project and the introspection project, were different cars where
used, showed quite different values on the x- and y-axis while driving on the
same route. Within the novice driver project it was also tried to put the BB in
the trunk which had also an effect on the data collection. A calibrated data
collection will also make the resulting data better comparable.
Recommendations concerning interpretation of the data
•
More test- and introspection rides have to be carried out in order to gain
more knowledge about the meaning of the data – i.e., what different
types of behaviour and interaction are reflected in the graphs produced by
the BB. And what are the critical values where one can say that they reflect
erroneous or dangerous behaviour. This applies for acceleration as well as
for the driven speed in curves.
•
The behaviour observation has to be better adapted to the content of the
test rides with the BB, as well. Many of the behaviour variables which are
observed and noted with the help of the Wiener Fahrprobe do not
necessarily have to be taken care of at this point of the project. Especially,
behaviour which has nothing to do with longitudinal and lateral acceleration
can be eliminated from the observation sheet. This means that for instance
the use of the indicator does not have to be recorded by the observers, in
this phase. On the other hand, better specified notes about speed
behaviour in relation to circumstances (acceleration, speed in curves) need
to be included.
5 Conclusion
In general it can be said that the behaviour observation and the data collection
method with a BB can easily be connected, and that they in principle can be
seen as complements of each others. The data of the acceleration on the three
axes have been visualised with the help of graphs and compared with the notes
of the behaviour observation. With the combined methods it was possible to
assess special types of behaviour like full braking at curves as well as other
sites could be identified and better explained and described.
Some difficulties occur when one only relies on BB-data and if one wants to
exactly determine the changes in the stimuli for the drivers, and to transform
such aspects into readable data that represent realistic conditions. No criteria
could be introduced that allowed to interpret exactly when a certain type of
acceleration on one of the axes is critical, meaning that an error or dangerous
behaviour occurred, like "really" driving too fast through curves or braking too
hard. So it was not possible at this stage of the project to say that one full
braking was more dangerous than another one. Only the statement that it was a
full braking could be made. To describe such mistakes no secure statements
can be given by only looking at the graphs. The connection with the behaviour
observation is needed to finally describe, and to understand, the behaviour of
the driver.
109
Human Centred Design for Intelligent Transport Systems
However, one of the future goals is that the graphs which display the BB data
should be used as a stand-alone in order to describe driving behaviour. The aim
could be to give driving instructors additional, concrete and demonstrative
material in order to assess and explain the driving style of a learner. On the
other hand it could be possible to assess the driving style of novice drivers and
give further recommendation about his/her driving style. Another possibility for
the use of such stand-alone graphs could be to assess the driving style of
drivers while driving with and without ITS in the car. As it was shown in the
results of this feasibility study two graphs could easily be compared and
different behaviour could be evaluated. This would also include the effects on
the driving style while using ITS.
To conclude, it has to be stated that the results of these two (pilot) projects are
very promising concerning the future goal of improving driver education by
"proving" with the help of BB data that erroneous or dangerous driving
behaviour really has occurred. However, to achieve this it will be necessary to
complete the BB data registration with other measurements, like, e.g. GPS and
XL-meter, and to add site descriptions to the acceleration data graphs. This will
help to give more detailed, more thorough, and better comparable descriptions
of the test rides.
6 References
[1]
[2]
[3]
[4]
[5]
[6]
Kaufmann C., Pípa M., Turetschek Ch. & Vašek J.: Black Box – Novice
Drivers & Black Box – Introspection, Final Report for HUMANIST TF2,
Vienna and Brno, 2007
Chaloupka Ch. & Risser R.: ‘Don't wait for accidents - possibilities to
assess risk in traffic by applying the Wiener Fahrprobe’, Safety Science 19,
Elsevier Science B.V., 1995
Risser R.: ‘Validity Problems in diagnostics and selection in’, in Risser R.
(Ed.): ‘Assessing the Driver. Faktor Mensch im Verkehr 41’ (Rot-Gelb-Grün
Braunschweig, 1997)
Risser, R., Lehner, U.: ‚Evaluation of an ACC (Autonomous cruise control)
system with the help of behaviour observation’. Presentation on the 4th.
ITS world-congress, 21-24. October 1997.
Turetschek, Ch.: ‘Effects of ISA-like IST - applications on the
communication behaviour of young car drivers in urban areas’, PhD Thesis
in progress
www.wien.gv.at, accessed February 2007
110
SESSION 3:
MODELLING OF DRIVERS’
BEHAVIOUR FOR ITS DESIGN
111
Modelling of drivers’ behaviour for ITS design
MODELLING DRIVER BEHAVIOUR IN ORDER TO
INFER THE INTENTION TO CHANGE LANES
Matthias J. Henning (Chemnitz University of Technology, Germany), Olivier
Georgeon (INRETS, France), Tony Wynn (SERI, United Kingdom), Josef F.
Krems (Chemnitz University of Technology, Germany)
matthias.henning@phil.tu-chemnitz.de
ABSTRACT: This study focuses on the examination and comparison of selected
behavioural and environmental indicators that predict the intention to change
lanes. These indicators were chosen from previous driving studies and driver
models. The data were gathered in a field study with an instrumented car that
can log data from the driver, the car, and the environment. The collected data
were analysed and modelled with the help of a "knowledge discovery
framework" (Georgeon, Mille, & Bellet, 2006). The first analysis of all lane
changes caused by a slow leading vehicle focuses on the following indicators:
glance to the left outside mirror, turn signal, and lane crossing. It is shown that
the glance to the left outside mirror could serve as a predictor with a high
potential to get information about the intention to change lanes in a very early
stage. However, it is important to combine this predictor with additional
information to avoid a high false alarm rate.
1 Introduction
The increase in the use of Advanced Driver Assistance Systems (ADAS) and
In-Vehicle Information Systems (IVIS) in recent years has created the need for
information about the driver and the traffic situation. To augment the benefit to
road safety and the driver’s acceptance of such systems, these ADAS and IVIS
should react according to the traffic situation and according to the driver’s
intention. For example, a Lane Departure Warning System (LDW) warns the
driver if he/she leaves the current lane. However, if the driver wants to execute
a lane change the system should not warn the driver.
A lane change is a ubiquitous driving manoeuvre in common driving
environments. It combines many critical features of driving such as low level
control, monitoring, and decision making. Lane change is defined by a more or
less stable sequence of actions which begins with a motivation to change lanes
(for instance a slow leading vehicle), followed by a gathering of information
about the surrounding traffic situation, and the decision whether to change
lanes or not, e.g. [1].
Several sources of data could be used to infer a driver’s intention to change
lanes. First there are car data; for example, using the turn signal could be a
valid predictor of the intention to change lanes. However, drivers do not use the
turn signal in every lane change situation. In a field study, Olsen [2] reported
that drivers tend to indicate a lane change in only 65% of all cases. Other
predictors could be steering wheel angle or acceleration [3]. A second source of
data is the car’s external environment [4]. The surrounding traffic, the lane
113
Human Centred Design for Intelligent Transport Systems
position, or geographical position can have some indicative power about the
intention to change lanes. Third, the driver himself could serve as a source. In
particular, eye movement behaviour could be a rich source of information about
the current goals and motives of the driver [5].
There have been several attempts to build algorithms to predict a lane change
manoeuvre. In a simulator study, Liu and Pentland [3] developed Hidden
Markov Models (HMMs) to predict a lane change to the left out of several other
driving manoeuvres. Prediction accuracy was 50 % 0.5 sec after the onset of
the manoeuvre, but increased in accuracy over time. Liu and Pentland [3] only
used information about the steering wheel for their models. Oliver and Pentland
[6] also used HMMs to predict a lane change manoeuvre to the left among other
manoeuvres in real traffic situations. Here, prediction accuracy was 23.5 % 0.1
sec before the manoeuvre took place. These authors used car data (speed,
acceleration, brake, gear, and steering angle) and gaze information. McCall,
Wipf, Trivedi, and Rao [7] used Sparse Bayesian Learning to develop an
algorithm to detect the intention to change lanes to the left. Information used in
this model was gas pedal position, brake pedal depression, longitudinal
acceleration, vehicle speed, steering angle, yaw rate, lateral acceleration, road
curvature metric, heading, lateral lane position up to 20 m ahead, and side-toside and up-down head movement. Real traffic data of only three participants
were used, and an accuracy of approximately 95 % was achieved, with 5 %
false alarms. Salvucci [8] proposed a “mind-tracking architecture” based on a
cognitive model of driver’s behaviour implemented in the ACT-R cognitive
architecture [9]. He reported a hit rate of 85 % with 4 % false alarms by using
information from the car, the environment, and the driver in a driving simulator.
All these attempts to build valid algorithms have some limitations. The studies
suffer either from a very small sample size or they were conducted in driving
simulators. Studies that were performed in real traffic environments with a
sufficient number of participants had a low prediction accuracy.
The goal of the present study is to focus on the impact of specific indicators and
their position in the lane change sequence. The present article describes in
more detail the timeline of the following indicators: slow leading vehicle, glance
to the left outside mirror, turn signal, and lane crossing. Those behavioural
indicators were chosen according to the Intent Detection Framework proposed
by Smith and Zhang [10].
2 Method
2.1
Participants
Data from 22 participants between the ages of 24 to 58 years were recorded
(MEAN = 33.8 years, SD = 10.1 years). 10 of them were female, and 12 of them
were male. Their driving experience in years ranged between 2 and 39 years
(MEAN = 13.4 years, SD = 9.7 years), with an annual quantum of driving from
2,000 to 50,000 km (MEAN = 13,136 km, SD = 10,508 km).
114
Modelling of drivers’ behaviour for ITS design
2.2
Instrumented car
A Renault Scénic was equipped to record and synchronize sensor data and
videos. The synchronization was done by using the time code from the video
frame. The logged sensor data came from car dynamics (speed, acceleration,
deceleration, yaw rate, and inclination), driver's behaviour (eye movement,
steering wheel position, pedal use, and turn signal), and environmental data
(distance to car ahead, GPS positioning). Video signals were recorded from five
sources: stereo-vision camera with radar for distance estimation to obstacles
(top left), the front view (top right), the rear view (down left – upper part), the
view from the left outside mirror down to the surface of the road (down left –
lower part), and the view to the participant’s head with the indications of the
eye-tracker (down right). Figure 1 shows the video logged during the
experiment.
Fig.1. Video sources from: stereo-vision camera with radar for distance
estimation to obstacles (top left), the front view (top right), the rear view (down
left – upper part), the view from the left outside mirror down to the surface of the
road (down left – lower part), and the view to the participants head with the
glance direction of the eye (down right)
2.3
Test course
The field study was conducted in the area of central France around the city of
Lyon. The subjects drove on a multi-lane motorway between Bron and the Lyon
International Airport (Saint Exupéry) in both directions. The total length of this
course was about 50 km. The speed limit varied between 90 km/h and 130
km/h.
2.4
Procedure
Participants were first informed about the goal of the study. They were told to
drive like they normally do. There was no information about the issue “lane
change manoeuvres” before and during the experiment. Then participants
received information about the car. The experimenter calibrated the eye tracker
115
Human Centred Design for Intelligent Transport Systems
and started the data logging. At least two experimenters were present during
the test drive that could influence the driving behaviour. However studies
indicate that the results are comparable to normal driving [11]. One of the
experimenter sat on the passenger’s seat and gave directions. The second one
sat in the back of the car, watched the driving scene, and indicated silently a
slow leading vehicle as a motive for a lane change by pressing a button, not
noticeably for the driver. This signal was logged with a time stamp and was
used for the post experiment interview. Participants were not urged to perform a
lane change manoeuvre at all. They could initiate it when they wanted to. After
the test driving, the data were immediately processed in order to find the
sequences with a motive to change lanes in the video. With this information
participants were questioned in two ways, depending on whether there was a
lane change following or not. If there had been no lane change following, the
participants were asked whether they had thought about executing a lane
change or not. In the case of thinking about executing a lane change they were
asked to show the starting point and the point where they had decided not to
change lanes in the video. If there had been a lane change following, the
participants were asked to show the point where they had started to think about
a lane change and the point after the execution of a lane change where they
had stopped thinking about the lane change. After the interview, participants
filled out a questionnaire on demographic data. Finally they were debriefed and
thanked.
2.5
Data Analysis
The following raw data was collected with the instrumented car: accelerator
pedal position, brake pedal position, clutch pedal position, indicator lights,
positions of gear stick, kilometric point, speed, steering wheel angle, distance to
object ahead, time to collision, headway, GPS longitude and latitude, GPS
indications about current road segment, and eye tracking data.
This raw data was analyzed through a step-by-step process of abstraction. This
process of abstraction was made with a methodology and a software tool called
ABSTRACT (Analysis of Behavior and Situation for menTal Representation
Assessment and Cognitive acTivity modelling) [12]. The implementation of this
tool took place in a collaboration between INRETS and the Chemnitz University
of Technology.
With this method, the analyst is able to specify “patterns of interest”, which are
relevant to the analysis of lane changes. These patterns of interest are:
acceleration, deceleration, stable speed, steering wheel angle threshold
crossing, steering wheel angle maximum, eye glance to left mirror, eye glance
to center mirror, eye glance to left side of the road, and obstacle detection.
Once specified by the analyst, these patterns are automatically computed. So
the overwhelming raw data is filtered in such a way that only the data that we
judge to be significant is finally displayed.
This computation produces a representation of the driving activity which is
shown in figures 3 and 4. In these figures, the circles at the bottom are events
on the basic level of abstraction (e.g. minima and maxima in the sensor data),
whereas the triangles and squares above are the symbols of the higher level
116
Modelling of drivers’ behaviour for ITS design
which are inferred from the basic level. Detailed descriptions of those symbols
are shown in the qualitative description of a lane change manoeuvre.
In a post experimental process, a coding of valuable indicators from the video
took place. The additional indicators were: distance to lane edge, distance to
the following car in the destination lane, and the size of the gap in the
destination lane. All these data were also incorporated into the ABSTRACT
software tool.
3 Results
3.1
Quantitative description of a lane change manoeuvre
Altogether, 194 lane changes to the left were analyzed. The duration between
the starting point of the lane change schema (‘start thinking about a lane
change’) and the actual lane crossing for all lane changes are shown in Figure
2.
frequencies
Histogram
cumulation %
40
100%
35
80%
frequencies
30
25
60%
20
40%
15
10
20%
5
0
0%
0
2
4
6
8
10
12
14
16
18
20
22
24
26
28
30 larger
class [sec]
Fig.2. Histogram of the duration of all lane change schemas
The mean value is 10.53 sec (SD = 10.18 sec) with a median of 7.82 sec. The
minimum was at 0 sec and the maximum at 88.64 sec.
All lane change manoeuvres were caused by a slow leading vehicle and were
performed in order to pass this vehicle. This study focuses on the following
indicators: a) first glance to the left outside mirror, b) turn signal, and c) the
actual lane crossing. There was at least one glance to the left outside mirror in
99.0 % of all lane changes and the turn signal was used in 99.0 %. In 87.1 % of
all lane changes the glance to the left outside mirror preceded the turn signal, in
11.3 % the order was preserved. Three lane changes were not preceded by a
glance to the left outside mirror or a signal or both. The mean duration between
the first glance to the left outside mirror and the lane crossing was 6.12 sec
(MEDIAN = 4.00 sec, SD = 7.01 sec). The mean duration between the onset of
the left turn signal and the lane crossing was 2.28 sec (MEDIAN = 2.00 sec,
SD = 2.17 sec).
117
Human Centred Design for Intelligent Transport Systems
3.2
Qualitative description of a lane change manoeuvre
As an example, two lane change sequences were modelled with the help of
ABSTRACT. The first schema (Figure 3) is characterized by the fact that it
begins in a situation, where the driver had to drive below his preferred speed. In
this schema, the acceleration associated glance into his left outside mirror
appeared as a good predictor of the lane change. One second after this
possible predictor the participant switched on the left turn-signal.
Fig.3. Lane change sequence with acceleration
In the second schema (Figure 4), the participant is not blocked by the obstacle
and he performs the lane change “on the fly”. In this case, there is no predictor
before the turn signal itself. Nevertheless, the turn signal appears to be a
sufficient predictor, since it is switched on by anticipation several seconds
before the manoeuvre.
Fig.4. Lane change sequence without any acceleration
4 Conclusions
The preliminary results of the analysis suggest that there is a potential for
several types of indicators and combinations of indicators to predict the
intention to change lanes. It is shown that the left turn signal and the glance to
the left outside mirror are two strong indicators. The high use of the turn signal
in this study seems due to the presence of the experimenter. Olsen [2] reports a
much lower frequency in using the turn signal in naturalistic driving conditions.
In addition, it is shown that the glance to the left outside mirror is of high
potential to increase the predictive power in two ways. On the one hand, the
glance to the left outside mirror allows an earlier prediction than the left turn
signal. On the other hand, the glance to the left outside mirror represents a
chance to predict the intention to change lanes even if the driver does not use
the turn signal. However, this glance to the left outside mirror is also present to
118
Modelling of drivers’ behaviour for ITS design
a certain extent if there is no intention to change lanes. In two studies, Henning
et al. [5] reported a probability for a lane change given a glance to the left
outside mirror of p = 0.40 and p = 0.79, respectively. It shows that 60 % and
21 %, respectively, of all glances to the left outside mirror are within the
baseline. The predictive power of the glance to the left outside mirror could be
increased with the help of other predictors like the acceleration which is shown
in figure 3. Another option could be the combination with the approaching of a
slow leading vehicle or the specific patterns of a driver. A detailed analysis and
discussion of this topic can be found in Henning and Krems [13].
This study has the chance to go beyond the previous work by a combination of
four features: i) the use of field data, ii) a sufficient number of participants, iii)
the chance to combine the indicators in a psychologically useful way, which
incorporates the knowledge about human processes in the field of attention,
memory, and decision making, and iv) the labelling of the lane change
sequence by three sources (experimenter in the car, observation and labelling
of the video after the driving by the participant, and a rating of three
independent observers on the basis of the video). The practical implication of
this study for the design of ADAS is that it shows that the use of eye glance
data and the incorporation of several sources of data from the environment
could not only increase the prediction accuracy, it also allows the assistant
system to have information about the driver’s intention earlier. One advantage
and at the same time a limitation is the presence of experimenters while driving.
This influences the driving behaviour to a certain degree [11] of the participants
so that it is not “naturalistic” anymore. To get an idea about the differences
between “naturalistic” and “observed” driving, it is planned to compare data from
naturalistic driving studies with the data of this study.
5 Acknowledgements
This project was granted by the European Union as part of the Network Of
Excellence (NOE) HUMANIST (HUMAN centred design for Information Society
Technologies applied to road transport).
6 References
[1]
[2]
[3]
[4]
McKnight, A. J., and Adams, B. B.: ‘Driver Education Task Analysis:
Volume I: TaskDescriptions’, Report No. DOT-HS-800-367, Washington,
DC: National Highway Traffic Safety Administration, 1970
Olsen, E. C. B.: ‘Modeling Slow Lead Vehicle Lane Changing’,
Dissertation, Virginia Polytechnic Institute and State University, Blacksburg
VA, 2003
Liu, A. and Pentland, A. P.: ‘Towards Real-Time Recognition of Driver
Intention’, Paper presented at the 1997 IEEE Intelligent Transportation
Systems Conference, Boston, MA., 1997
Baker, M., and Carter. M.: ‘Development of a Lane Changing Algorithm for
Evaluating the Impact of ICC’, Report, Science Applications International
Corporation, McLean, VA, for Volpe National Transportation Systems
Center, Cambridge, MA, 1997
119
Human Centred Design for Intelligent Transport Systems
[5]
[6]
[7]
[8]
[9]
[10]
[11]
[12]
[13]
Henning, M. J., Schweigert., M., Baumann, M., and Krems, J. F.: Eyeglance patterns during lane change manoeuvres’, Paper presented at the
11th conference on Vision in Vehicles, Dublin, Ireland, 2004
Oliver, N., and Pentland, A.: ‘Graphical models for driver behavior
recognition in a smartcar’, Proceedings of the IEEE lntelligent Vehicles
Symposium, 2004
McCall, J., Wipf, D., Trivedi, M., and Rao, B. D.: ‘Lane Change Intent
Analysis Using Robust Operators and Sparse Bayesian Learning’,
Technical Report, San Diego La Jolla: University of California, 2004
Salvucci, D. D.: ‘Inferring driver intent: A case study in lane-change
detection’, Proceedings of the Human Factors Ergonomics Society 48th
Annual Meeting, 2004
Anderson, J. R., and Lebiere. C.: ‘The atomic components of thought’,
Hillsdale, NJ: Erlbaum, 1998.
Smith, M., and Zhang, H.: ‘SAVE-IT - SAfety VEhicles using adaptive
Interface Technology (Task 9): Intent Inference’, Washington D.C.:
National Highway Traffic Safety Administration, 2004
Hjälmdahl, M., Várhelyi, A. ‘Validation of in-car observations, a method for
driver assessment’, Transportation Research Part A, Vol 38, no 2, pp127142, 2004
Georgeon, O., Mille, A., and Bellet, B. T.: ‘Analyzing behavioral data for
refining cognitive models of operator’, Paper presented at the 17th
International Conference on Database and Expert Systems Applications
(DEXA'06), Krakow, Poland, 2006
Henning, M. J., and Krems, J. F.: ‘The impact of behavioural indicators to
infer the lane change intention’, Paper in preparation for publication, 2007
120
Modelling of drivers’ behaviour for ITS design
ANALYSIS AND MODELLING OF DRIVER
PREPARATORY BEHAVIOUR BEFORE TURNING
AT INTERSECTIONS
Toshihisa Sato, Motoyuki Akamatsu (AIST, Japan)
toshihisa-sato@aist.go.jp
ABSTRACT: This study focuses on an analysis and modelling of the naturalistic
driving behaviour before making a right turn at an intersection. We developed
instrumented vehicles and conducted long-term experiments on a public road to
measure driver behaviour, vehicle state, and headway and rear distances. The
relationships between the onset locations of covering the brake pedal and
activating the turn signal and the traffic conditions of driving with or without lead
and/or following vehicles were analyzed based on the data measured in the real
road environment. The results suggest that the existence of the front and rear
vehicles and the vehicle velocity influence the onset location of covering the
brake pedal. Structural equation modelling was applied to estimate these
relationships quantitatively. The results imply that the model with two latent
variables, free-driving condition level and location of transition to preparatory
behaviour, can represent the hypotheses obtained from the naturalistic
behaviour analysis. Finally, we investigated the driver preparations before
making turns at other intersections with various road structures and discussed
the differences of the influence of the traffic conditions between the road
environments.
1 Introduction
The popularization of in-vehicle navigation and telecommunication systems is
progressing rapidly all over the world as one of key components for realizing
ITS. Drivers use route guidance information via the in-vehicle navigation
systems while driving from the origin to the destination. The route guidance
instruction helps drivers to choose and maintain the correct routes with lower
mental workload. In this study, we focused on the presentation timing of the
voice instruction just prior to making the turn. This information can trigger a
change in driving behaviour from straight mode to preparation mode while
approaching a target intersection. The presentation timing is usually constant in
any road traffic conditions. However, the constant timing can lead to a reduction
in the driver acceptance of the provided information, because there may be
some variations in the driving operations due to the influence by the road traffic
environment [1]. It is essential to develop a human-centred design, i.e. to
develop a presentation timing adapted to the typical driver’s preparatory
behaviour before making a turn at an intersection based on the investigation on
the naturalistic driving behaviour under actual road traffic environments.
We conducted long-term experiments using instrumented vehicles on a public
road in order to measure the natural driving behaviour before making a right
turn at a specific intersection. This experiment was carried out in left-hand
121
Human Centred Design for Intelligent Transport Systems
driving, i.e. under Japanese traffic contexts. We focused on leading and
following vehicles as the traffic conditions in the vicinity of the target
intersection. In this study, the leading vehicle is defined as a forward vehicle
that travels straight toward the target intersection. We did not analyze the traffic
situations in which the forward vehicle turned to the right, as did the driver’s own
vehicle, because the stopping locations for waiting for the right turn when there
are leading vehicles stopping at the target intersection are different from those
without the leading vehicles. The following vehicle is defined as a vehicle that
follows the driver’s vehicle in the same traffic lane, regardless of whether or not
the following vehicle turns to the right at the turning point.
1.1
Aims of this study
(a) Analyze naturalistic driving behaviour before making a right turn at a specific
intersection and investigate the influence of the traffic conditions in the vicinity
of the target intersection on the onset location of the driver’s preparatory
behaviour.
(b) Construct a driver model for describing the relationships between the traffic
conditions and the driver behaviour.
(c) Investigate the driver’s preparatory behaviour before making turns at other
intersections with various road structures in order to clarify the road conditions
where the driver’s behaviour is influenced by the existence of the leading and
following vehicles.
2 Experiments
2.1
Participants
Four non-professional drivers (three males and one female) participated in the
long-term experiments. The average age of the participants was 34.8 years
(age range: 22 - 52 years). The average driving experience was 16.3 years
(experience range: 3 - 33 years). All participants drove a passenger vehicle
almost every day in their daily lives.
2.2
Experimental vehicle
Figure 1 presents an instrumented vehicle to measure driver behaviour used in
this experiment. Various sensors and a recorder system detected the vehicle
driving state, including vehicle velocity, vehicle acceleration, and geographical
position, and measured driver’s behaviour, including steering, accelerating, and
braking operations. The relative distance and relative speed of the lead and
following vehicles were measured with two laser radar units attached to the
front and rear bumpers. The driver’s foot position (covering the accelerator
pedal or brake pedal without pressing) was detected by laser sensors fitted
above the pedal surfaces. The turn signal activation was detected by adding
encoders to the lever. The data was recorded on a laptop computer and mobile
hard disks via a driving recorder system. This recorder system was fixed inside
the trunk of the instrumented vehicle to encourage naturalistic driving behaviour
of the participants.
122
Modelling of drivers’ behaviour for ITS design
Driving recorder system
CCD camera for
recording left line
D-GPS sensor
microphones
Laser
radar
Laser
radar
CCD cameras
Steering sensor
Lever sensor
Pedal sensors
Speed sensor
Laptop PC
CCD camera for
recording right line
G-sensor and gyro sensor
Fig.1. Overview of the instrumented vehicle
2.3
Target intersection
We carried out repeated experiments on a public road in Tsukuba. The selected
driving route was a 30-minute trip (total mileage: about 15 km) that included
several left and right turns. Figure 2 presents a diagram and images of the
analyzed intersection on the experiment route. The intersection has a
designated lane for making a right turn. There is a long straight road about 2 km
as far as the target intersection, which has two traffic lanes.
Centre of intersection
70m
Fig.2. Overview of the target intersection
123
Human Centred Design for Intelligent Transport Systems
2.4
Variables
Driver’s preparatory behaviour before making the turn corresponds to activating
the turn signal and decelerating. We focused on the foot movement to cover the
brake pedal as the decelerating operations.
The behavioural events were quantified by using the remaining distance to the
centre of the target intersection (the stop line for turning to the right) at the onset
of each preparatory manoeuvre. In an analysis of the driver preparatory
behaviour, we investigated the onset location of each operation by classifying
the traffic conditions while approaching the target intersection, i.e. driving with
both leading and following vehicles, driving with a leading vehicle and without a
following vehicle, driving without a leading vehicle and with a following vehicle,
and driving without either a leading or following vehicle, in order to clarify
general influence of the traffic conditions on the preparatory behaviour. The
driving with a leading or following vehicle was defined as the existence of a
vehicle running in front of or to rear of the driver’s own vehicle before the driver
entered the designated right-turn lane. In a construction of the driver model, we
used the vehicle velocity when the brake pedal was covered by the driver’s right
foot, the remaining distances at the onset of each preparatory behaviour, and
the relative distances to the leading and following vehicles before the driver
entered the designated right-turn lane, in order to evaluate in detail the impacts
of the traffic situations on the preparatory manoeuvres.
2.5
Experimental procedure
We developed the four instrumented vehicles, and four participants started
driving on the identical route at 10-min intervals to measure each participant’s
driving behaviour under similar traffic conditions. The recorded trip was made
once a day on weekdays and the total was 40 trips (over a period of about eight
weeks). Practice drives were made before the measurement trials so that the
participants could drive from the origin to the destination without using a map.
The participants rode alone in the instrumented vehicle during the experiment
trials. They were instructed to drive in their typical manner.
3 Analysis of driver preparatory behaviour
The average remaining distance to the centre of the target intersection when
each behavioural event occurred was calculated in the categorized traffic
conditions. Figure 3 presents the results of the relationships between the onset
location of the driver preparatory behaviour and the traffic conditions.
The remaining distances to the centre of the target intersection when covering
the brake pedal without either a leading or following vehicle were the longest
among the four traffic conditions. The closest onset locations of the right foot
movement were found when driving with both leading and following vehicles.
The existence and number of other vehicles around the driver’s vehicle may
influence the onset location of driver’s decelerating manoeuvre.
The remaining distances when activating the turn signal with leading and
following vehicles were shorter than those during trials without leading and
124
Modelling of drivers’ behaviour for ITS design
following vehicles. However, the decreasing tendencies were not remarkable
compared to the results of covering the brake pedal. The onset location of the
turn signal activation ranged from 80m to 95m and was independent of the
existence of other vehicles in front of or to rear of the driver’s vehicle.
160
(m)
160
140
140
120
120
100
100
80
80
60
60
40
40
20
20
0
With leading With a leading
Without leading
and following
vehicle
and following
With a following
vehicles
vehicles
vehicle
(a) Cover a brake pedal
0
(m)
With leading With a leading
Without leading
and following
and following
vehicle
With a following
vehicles
vehicles
vehicle
(b) Activate a turn signal
Fig.3. Average onset location of each behavioural operation in the four traffic
conditions
It is hypothesized that the drivers drive at a lower speed when they follow a
forward vehicle. Figure 4 confirms this hypothesis. The driving speeds while
driving with leading and following vehicles and with a lead vehicle were lower
than those of the driving conditions without a leading vehicle. The participants
drove the fastest when the leading and following vehicles did not drive in front of
and to rear of the participants’ vehicles.
In the comparison analyses of the driver’s preparatory behaviour in a restricted
vehicle velocity from 16m/sec to 17.5m/sec, it was suggested that the locations
at the onset of covering the brake pedal were closer to the centre of the
intersection when driving with leading and following vehicles, compared to
driving without a leading vehicle.
25
(m/sec)
20
15
10
5
0
With leading With a leading
Without leading
and following
vehicle
and following
With a following
vehicles
vehicles
vehicle
Fig.4. Average driving speed when covering the brake pedal in the four traffic
conditions
125
Human Centred Design for Intelligent Transport Systems
4 Modelling of driver preparatory behaviour using structural
equation model
4.1
Method and result
The behavioural analysis suggests that the existences of leading and/or
following vehicle correlate with the vehicle velocity while approaching the target
intersection. The experiment results reveal a linear relation between the onset
location of driver’s preparatory behaviour and the traffic conditions. We evaluate
these relationships quantitatively by using the structural equation model with the
following five variables: the vehicle velocity, the relative distance to the leading
vehicle, the relative distance to the following vehicle, the remaining distance to
the centre of the target intersection when covering the brake pedal, and the
remaining distance to the centre of the target intersection when activating the
turn signal. When there are no leading and following vehicles while approaching
the target intersection, the relative distances to the forward and rear vehicles
were compensated with the driving speed based on a regression equation
between the driving speed and the headway or rear distance under conditions
of driving with leading or following vehicles.
Figure 5 presents a path diagram of the proposed structural equation model
after several trials of model construction and estimation, and the results of
model fit indices. We introduced two latent variables, free-driving condition level
describing relative distances to leading and following vehicles and location of
transition to preparatory behaviour describing onset locations when covering the
brake pedal and activating the turn signal.
Vehicle
velocity
R2=0.944
0.50
0.97
R2=0.860
0.97
Relative distance
to leading vehicle
R2=0.590
E4
Relative distance
to following vehicle
R2=0.336
R2=0.933
0.77
0.43
Free-driving
condition level
2
0.58
D2
RD when activating
turn signal
E2
E1-E4: Measurement error D1,D2: Prediction error
0.77
RD: Remaining Distance
D1
χ
E1
Location of transition to
preparatory behaviour
R2=0.592
E3
RD when covering
brake pedal
(df)：3.302(3) p value：0.347
GFI：0.990
AGFI：0.949
CFI：0.999 RMSEA：0.028
GFI: Goodness-of-Fit Index AGFI: Adjusted Goodness-of-Fit Index CFI: Comparative Fit Index
RMSEA: Root Mean Square Error of Approximation
Fig.5. Path diagram of the constructed structural equation model
The chi-square value is non-significant, indicating an acceptable model-to-data
fit. In addition, the results of the model fit indices demonstrate that the proposed
structural equation model fits the observed data well. The path coefficients
between the vehicle velocity and the location of transition to preparatory
behaviour and between the free-driving condition level and the location of
transition are almost equal, suggesting that the influence on the onset locations
of the preparatory behaviour is almost equal between from the vehicle velocity
126
Modelling of drivers’ behaviour for ITS design
and from the relative distances to the leading and following vehicles. The factor
loadings from the location of transition to preparatory behaviour to each
preparatory manoeuvre present the difference in the impacts of the driving
speed and the traffic conditions on each onset location, i.e. the two factors have
large effect on the onset location of covering the brake pedal and have less
effect on the location at the onset of activating the turn signal, corresponding to
the analysis results of the driver preparatory behaviour.
4.2
Implication for route guidance presentation
The constructed structural equation model suggests that the driving speed and
the relative distances to lead and following vehicles should be taken into
account when determining the presentation criteria of the route guidance
instruction. This is because the driver’s onset location where they begin to
decelerate before making the turn differs according to the vehicle velocity and
the traffic conditions while approaching the target intersection. For example, the
presentation timing can be closer to the target intersection when drivers drive
slowly or they follow a lead vehicle.
In addition, the proposed structural equation model may contribute to predicting
the onset locations of the driver’s foot movement to cover the brake pedal and
turn signal activation, based on the vehicle velocity and the relative distances to
leading and following vehicles when the route guidance information is provided
for drivers. We confirm higher prediction accuracy of the proposed structural
model, compared to the prediction by a single regression model of vehicle
velocity, by evaluating the differences between the predicted values and the
observed values at the onset of each behavioural event which were not used for
the model construction. The in-vehicle system linked with the speed pulse
sensor and laser radar units can predict the driver’s onset location of
preparation for a right turn by using the path coefficients of the constructed
structural model. If driver does not begin to prepare to make a right turn after
reaching the predicted onset location, the system can access the driver’s
navigational error, i.e. the driver did not accurately identify the turning point, or
did not notice the information provision, and then restate the route guidance or
issue a warning.
5 Application to other intersections on other manoeuvres
and with various road structures
We focus on only one specific intersection with two traffic lanes and a
designated right-turn lane. We analyse the driver’s preparatory behaviour while
approaching the other intersections with various road structures and on the
other manoeuvres involved with a left turn, and investigate the influence of the
existence of the leading and following vehicles on the onset location of the
driver’s decelerating manoeuvre.
Figure 6 presents an overview of the analyzed intersections on the same
experimental route as the target intersection in Fig. 2. Intersection 2 has two
traffic lanes, and the other intersections have one traffic lane. The participants
127
Human Centred Design for Intelligent Transport Systems
made a left turn at Intersection 2 and Intersection 6. They approached
Intersection 4 and Intersection 6 after driving along curves.
Figure 7 presents the results of the onset locations of covering the brake pedal
in the four categorized traffic conditions at each intersection. At Intersection 1,
Intersection 2, and Intersection 5, the remaining distances to the centre of the
intersections while driving with leading and following vehicles were shorter
compared to the drives without either a forward or rear vehicle, indicating the
similar tendencies to the analysis at the specific intersection with a designated
right-turn lane.
At Intersection 4 and Intersection 6, the onset locations of the root movement to
cover the brake pedal while driving with the leading and following vehicles were
the longest among the four traffic conditions. The drivers tended to begin to
prepare to make a turn at an earlier point while they approach the intersection
after a curve in a car-following condition compared to in a free-driving condition.
At Intersection 3, the onset location of covering the brake pedal is closer to the
centre of the intersection while driving with only a leading vehicle or without
leading and following vehicles, compared to driving with a following vehicle. A
road construction was conducted around Intersection 3, and the participants
tended to begin to decelerate earlier while approaching the intersection with a
following vehicle, due to avoidance of the rear-end collision caused by the
following driver who is distracted from noticing the intersection.
The results of the additional analyses suggest that the same relationships
between the driver’s preparatory behaviour and the traffic conditions as the
specific intersection with a designated right-turn lane are found on the left-turn
manoeuvre and the intersection with one traffic lane where there is no
designated lane for making a turn. However, the relationships are different on
the intersection with a road construction and the intersection after a curve. It is
important to take into consideration the traffic conditions in a different manner
when determining the route guidance presentation criteria at the intersections
with a curve or a road construction.
128
Modelling of drivers’ behaviour for ITS design
GOAL
Intersection 5
(Right turn with one traffic lane)
Intersection 6
(Left turn after a curve)
Intersection 3
(Right turn along road construction)
Intersection 4
(Right turn after a curve)
Intersection 1
START
Intersection 2
(Right turn with one traffic lane)
(Left turn with two traffic lanes)
Fig.6. Map of the experimental route and analyzed intersections
6 Conclusion
We developed instrumented vehicles and conducted experiments on a public
road to measure driver’s preparatory behaviour while approaching intersections,
vehicle velocity, and traffic conditions around the driver’s vehicle. The
relationships between the onset locations of covering the brake pedal and
activating the turn signal and the existence of the forward and following vehicles
were analyzed based on the data measured at a specific intersection with a
designated right-turn lane. The results suggest that the existence of the front
and rear vehicles and the vehicle velocity influence the onset location of
covering the brake pedal. Structural equation modelling was applied to estimate
these relationships quantitatively. The results imply that the model with two
latent variables can represent well the hypotheses obtained from the naturalistic
behaviour analysis.
Finally, we investigated the driver preparations before making turns at other
intersections on the other manoeuvre and with various road structures. The
additional investigation results indicate that the traffic conditions around the
driver’s vehicle and the driving speed have an influence on the onset location of
the driver’s decelerating operation before making a left turn or while driving on
one traffic lane, and the influence is the same as the specific intersection with
the designated lane. However, the influence differs at the intersections with a
road construction and the intersection after a curve. Further studies with various
categories of drivers will be conducted to validate the findings obtained from this
analysis.
129
Human Centred Design for Intelligent Transport Systems
(m)
100
(m)
100
80
80
60
60
40
40
20
20
0
0
With leading With a leading
Without leading
and following
vehicle
and following
With a following
vehicles
vehicles
vehicle
With leading With a leading
Without leading
and following
and following
vehicle
With a following
vehicles
vehicles
vehicle
Intersection 1 (right turn)
Intersection 5 (right turn)
(a) Standard Intersections with one traffic lane
(m)
70
60
60
(m)
50
50
40
40
30
30
20
20
10
10
0
0
With leading With a leading
Without leading
and following
and following
vehicle
With a following
vehicles
vehicles
vehicle
With leading With a leading
Without leading
and following
and following
vehicle
With a following
vehicles
vehicles
vehicle
Intersection 4 (right turn)
Intersection 6 (left turn)
(b) Intersections after a curve
140
(m)
(m)
100
120
100
80
80
60
60
40
40
20
20
0
With leading With a leading
Without leading
and following
and following
vehicle
With a following
vehicles
vehicles
vehicle
Intersection 2
(left turn with two traffic lanes)
0
With leading With a leading
Without leading
and following
vehicle
and following
With a following
vehicles
vehicles
vehicle
Intersection 3
(right turn along road construction)
(c) Other Intersections
Fig.7. Average onset locations of covering the brake pedal on the analyzed
intersections
7 References
[1]
Yoshioka, M., Akamatsu, M. Matsuoka, K. and Sato, T. ‘Aim of behaviorbased human environment creation technology’. Proc. of the XVth
Triennial Congress of the International Ergonomics Association, 2003, CDROM
130
Modelling of drivers’ behaviour for ITS design
INFLUENCE OF UNEXPECTED EVENTS ON
DRIVING BEHAVIOUR AT DIFFERENT
HIERARCHICAL LEVELS: A DRIVING SIMULATOR
EXPERIMENT
T. W. Schaap (Twente University, The Netherlands), A. R. A. van der Horst
(TNO, The Netherlands), B. van Arem (Twente University, The Netherlands)
t.w.schaap@utwente.nl
ABSTRACT: Computer based simulation models of human driving behaviour
can be used effectively to model driving behaviour and behavioural adaptation
to Intelligent Transport System (ITS). This can be a useful step in human
centered design of ITS. To construct a comprehensive model of driving
behaviour, the interaction between the three levels of the driving task has to be
determined. This gives insight into how different driving tasks influence each
other. A driving simulator experiment was conducted to determine the
relationship between levels of the driving task. The influence of workload on this
relationship was determined by giving subjects an additional cognitive task.
Subjects had to drive many similar intersections, with two unexpected events.
Their reaction on the tactical level to the compensation on the control level was
measured. Participants lowered speed and increased headway after having to
brake; level of unexpectedness increased this effect. Workload decreased this
effect on driving speed.
1 Introduction
1.1
Modelling behavioural adaptation
Transportation Systems (ITS)
to
Intelligent
Driving is and has always been a complex task. Next to controlling the vehicle,
which is a complex task in itself, the driver has to determine why, when and
where he wants to drive. He has to determine his route, make sure he takes the
correct turns, avoids collisions with vehicles crossing his lane and keeps the
appropriate distance to lead vehicles. Furthermore, many secondary tasks such
as tuning the radio or talking on the phone are performed while driving. This
requires constant attention, goal management and use of memory. Today, in a
time with more vehicles on the road than ever [1], both industry and
governments have set their focus on making driving safer and more
comfortable. One of the ways to achieve this goal is through the development of
Intelligent Transport Systems (ITS).
When developing ITS, the driver is not always at the center of the designing
process. Therefore, it is not always guaranteed that the application will fulfill a
certain user need. Furthermore, in many cases it is uncertain how the human
user of ITS will react to the system. This is an important step in the designing
process of ITS. Drivers may adapt their behaviour in an unexpected way, which
131
Human Centred Design for Intelligent Transport Systems
may turn out to be both positive and negative. Determining the effects of this
behavioural adaptation can be done in many ways, depending on the stage of
the development process and the possible safety effects of the system [2]. A
cost-effective and safe way to determine the effects of ITS is by developing a
computer simulation model of driving behaviour, which can interact with such a
system. We will develop a simulation model of driving behaviour based on the
structure of the driving task, making it possible to develop a comprehensive
driving behaviour model.
1.2
Levels of the driving task
Our computer simulation model of driving behaviour will have to meet a number
of constraints [2], such as running in real-time, focusing on intersection
behavior, and the possibility to build the model structure on Michon’s
hierarchical model of the driving task [3]. According to Michon [3], driving
consists of three hierarchically ordered levels: strategic, tactical and operational.
On the strategic (navigation) level, the goals of the trip are set and the route and
departure time are determined. On the tactical (guidance) level, the driver has
to follow the road, maintain a steady speed and keep enough distance to other
vehicles. On the operational (control) level, the driver controls the vehicle by
pressing the gas pedal and the brakes, turning the steering wheel and using the
vehicle controls. These levels can be active at the same time and can influence
each other. This is most clear at intersections, where route choice, tactical
maneuvers and control tasks are equally important. This is also where the topdown influence between the levels becomes most clear. When a certain route
has been chosen, the driver has to make the turns according to this route and
therefore control the vehicle to do so.
On the other hand, bottom-up influence is also possible. Michon [3] expected
that this would have a relation with expectations and unexpected situations.
Unexpected situations can lead to a higher priority for lower level tasks when
they have to take over to guarantee safety or the original task can not be
performed successfully. For instance, when one drives a certain route and a
street turns out to be blocked, the tactical level solution of taking a different turn
will take precedence over the original route. When a driver suddenly has to
brake for a crossing child, the distance to other vehicles or the next turn to take
can be altered according to the outcome of the situation. In these cases, lower
levels influence the outcome of higher levels.
Alexander and Lunenfeld [4] describe how primacy increases with lower level
tasks, whereas complexity increases with higher level tasks. Tasks on a higher
level therefore often take more time to complete and are more complex. On the
other hand, lower level tasks sometimes have to take over in order to ensure
safety. These latter situations are of interest for our research, because this is
bottom-up influence between levels of the driving task. These are the
situations in which the driver compensates for unexpected events.
Unfortunately, the hierarchical relation between the levels of the driving task,
and especially the bottom-up influence resulting from this, has not yet been
determined precisely, and can therefore not yet be fully integrated in a
computational model of driving behaviour. Only when we know how normal
132
Modelling of drivers’ behaviour for ITS design
driving behaviour takes place, we can make a valid and complete
comprehensive model of driving behaviour, which can in turn be used to
determine the effects of behavioural adaptation to ITS.
1.3
Research objective
We conducted a driving simulator experiment to determine in which situations
normal top-down interaction between the levels of the driving task is overruled
by bottom-up influence. This bottom-up influence can be seen as compensation
for an unexpected event. In order to determine the relation between this event
and the participants’ reactions, we also looked at the influence of cognitive
workload and at the level of unexpectedness of a situation on this bottom-up
influence. We focused on the tactical level and the control level, because it is
very difficult in an experimental setting to control the expectations of a group of
participants on the strategic level (route and trip goals).
1.4
Hypotheses
An unexpected event will cause compensation behaviour on a lower level of the
driving task, influencing higher level tasks. After a while, the effect of the
unexpected event on the task will fade away. The operational level tasks that
we studied were sudden braking as a reaction to a braking lead vehicle, and
steering away from an approaching vehicle from the left. These operational
level tasks will have an impact on following distance and intersection approach
speed, as well as anticipation and lateral acceleration, which are tactical level
tasks.
We expect that compensation behaviour on the operational level to a braking
lead car (by braking), increases the participants’ following distance, increases
anticipation to the intersection (measured by minimum speed based on distance
and time to intersection) and decreases overall intersection approach speed.
After a number of intersections, this effect will fade away. We also expect that
the level of unexpectedness has an influence on this.
We furthermore expect that compensation behaviour in reaction to an
accelerating vehicle from the left decreases overall intersection approach speed
as well as the distance-to-intersection and time-to-intersection of the onset of
speed decrease (anticipation), and influences lateral acceleration. After a
number of intersections, this effect will fade away. We also expect that the level
of unexpectedness has an influence on this.
2 Method
2.1
Participants
87 subjects participated in our experiment. They were between 23 and 60 years
old, had their driver’s license for five years or more and drove 10.000 kilometers
or more annually. Due to simulator sickness, 11 participants did not complete
the experiment. 76 participants completed the experiments. An error in the data
storage led to an incomplete dataset for 37 participants. 39 complete datasets
were used for analysis.
133
Human Centred Design for Intelligent Transport Systems
2.2
Data measurement
The experiment was performed in a driving simulator of TNO Human Factors.
This simulator was a fixed-based driving simulator with manual transmission
(see Figure 1). The participant could control the driving simulator by means of
normal vehicle controls. The road environment and other road users were
projected on three screens with a total horizontal field of view of 180° and the
total vertical field of view of 45°.
In order to determine the participants’ workload during driving, participants had
to perform a Peripheral Detection Task (PDT) [5]. Measuring workload
objectively is difficult with self-report measures [6], but with the PDT, this is less
of a problem. A LED light is shown to the participants randomly every 3 to 5
seconds (see Figure 1). Participants have a small switch attached to their index
finger, which they have to press every time they see the LED light. Workload is
determined by reaction times and the number of missed signals. As workload
increases, visual attention narrows [5], which increases response time and the
chance of missing a signal. If a participant did not respond to the LED light for 2
seconds or more, this was registered as a missed signal.
2.3
Experimental design
An urban layout was simulated with one long road, crossed by twenty other
roads, creating 20 intersections. Subjects were instructed to drive with a
maximum speed of 50 km/h and give priority to traffic coming from the right,
according to Dutch traffic regulations. They had to go straight on each
intersection and park the simulated car at the end of the road (see Figure 1).
The infrastructure was similar at each intersection, and the other road users
always behaved similarly. A lead car drove in front of the subject the whole
time. This vehicle slowed down when the subject fell too far behind, and
speeded up when the subject came too close. This way, the lead distance was
always between 18 and 48 meters from the bumper of the simulated car.
When the participant was approaching an intersection, a car coming from the
right always crossed the intersection first, followed by a car from the opposite
direction. The participant and a lead car then reached the intersection, and a
car from the left yielded and crossed after they had passed the intersection. The
intersection layout and the positioning of other road users, including the lead
car, are depicted in Figure 1.
The experiment consisted of four experimental drives and an introduction drive.
In the first experimental drive, participants only drove on standard intersections,
setting their expectations about the situations to come. This was the reference
condition for normal driving. After 5 standard intersections in the second drive,
an unexpected event happened: the lead vehicle suddenly braked. The third
drive was a reference condition again. At the eighth intersection in the fourth
drive, the car from the left that had always yielded now suddenly accelerated
and stopped before colliding with the participants. Both unexpected events
could occur at three levels: strong, medium and mild.
134
Modelling of drivers’ behaviour for ITS design
Fig.1. Experimental setup
(1a: intersection layout with traffic; 1b: lead car; 1c: LED light for PDT; 1d:
switch for PDT; 1e: end of experimental drive; 1f: the TNO driving simulator)
Half of the participants were furthermore given an additional, cognitive task, to
determine the effect of cognitive load on the influence between the levels of the
driving task. They had to count back from a high number by steps of a certain
size (ranging between 4 and 9, depending on the drive). All participants with the
additional task were given exactly the same task, and they were told to do it as
fast as possible, making sure that they had additional cognitive load during
driving. Participants were encouraged to continuously perform the additional
task and were reminded of their task after a number of seconds without an
answer, but they could give their final answers at their own pace, ensuring that
all participants were equally challenged.
The two levels of cognitive workload (no additional task or additional task) and
the three levels of unexpectedness of the events (mild, medium, hard) leads to
6 event conditions. The event conditions were counterbalanced among
participants to eliminate possible learning effects. Participants always
encountered the same version (mild/ medium/ hard) of both the unexpected
events, and always drove a drive without an unexpected event as the first and
third drive. Participants with an additional cognitive task also had to perform this
cognitive task during reference conditions.
After each drive, participants were given a break and a questionnaire, with
questions related to simulator sickness, the level of predictability of the driving
task and the level of difficulty of the PDT. For the participants with the counting
task, additional questions were asked about the level of difficulty of this task,
and whether it had influenced the participant’s way of driving.
135
Human Centred Design for Intelligent Transport Systems
2.4
Data registration
Twenty-one variables were registered with a frequency of 256 Hz during the
experiment:
1. Time (s);
2. Path number;
3. Distance to intersection (m);
4. Distance from next intersection (m);
5. Velocity (m/s);
6. Acceleration (m/s2);
7. Lateral position (m);
8. Lateral velocity (m/s);
9. Steering angle (angle);
10. Gas pedal angle (percentage of maximum pressed);
11. Brake pedal angle (percentage of maximum pressed);
12. Time headway (s);
13. Distance lead vehicle (m);
14. Velocity lead vehicle (m/s);
15. Time to intersection (s);
16. Time to collision (s);
17. Time to collision with left approaching vehicle (s);
18. Time to collision with right approaching vehicle (s);
19. Distance to intersection of lead vehicle (m);
20. Speed of lead vehicle (m/s);
21. PDT reaction time (ms);
2.5
Analysis
All recorded twenty-one variables are related to tactical level or control level
driving tasks. These were used to answer the question whether, and in what
way, control level compensation for an unexpected event influences tactical
level task performance. In this paper, we only focus on the first unexpected
event, the braking lead car. A full description of the second part of the analysis
and its results will be published in [7].
First, we determined whether the standard intersections were actually seen by
the participants as "expected" and the unexpected events as “unexpected”. This
was determined by studying the answers given in the questionnaire and by
looking at learning effects in driving. Next, we tested our hypotheses that the
braking lead car would have an effect on following distance, moment and
136
Modelling of drivers’ behaviour for ITS design
location of anticipation to the intersection and the intersection approach speed.
This was tested by comparing these variables in the reference case (third drive)
to the intersections directly after the event. Finally, the effects of the level of
unexpectedness (mild, medium or hard braking) and of the additional cognitive
task were examined.
3 Results
3.1
Expectations
The expectations of the participants were tested by studying their driving
behaviour and related learning effects, and by examining their answers to
selected questions of the questionnaire.
3.1.1 Learning effects in driving behaviour
A learning effect can be determined by examining the average speed on
intersections. Average driving speed increases with experience, as a result of
the participants’ knowledge of what to expect. Indeed, a significant increase of
average driving speed can be found during the experiment (p<.0001).
The standard deviation of speed also tells us something about the learning
effect of standard intersections: when participants get more used to situations,
they know better what to expect and therefore can drive more smoothly trough
these intersections. A decrease in the standard deviation of speed therefore
also points to a learning effect. The standard deviation of speed decreased
significantly during the experiment (p<.0001). This supports our expectation that
participants were getting used to the standard intersections during the first
experimental drive.
3.1.2 Answers to selected questions of the questionnaire
Participants were given a questionnaire after each experimental drive. Some
questions concerned the participant's comfort during the experiment, some
were related to the difficulty of the driving task, the PDT and the additional task.
A third topic in the questionnaire was the predictability of other road users’
behaviour in the experiment and the intersections. The answers to these
questions were used to determine whether participants had expectations about
the situations and whether an unexpected event was actually seen as
unexpected. The answers to the questionnaire revealed that the reference
conditions were seen as significantly more predictable than the drives with the
unexpected events (p<.006, see Figure 2). Also, 50% of the unexpected events
were explicitly mentioned in the field for additional information about the
experiment. It is safe to conclude that the unexpected events were really not
expected by the participants.
137
Human Centred Design for Intelligent Transport Systems
Fig.2. Level of unexpectedness per drive
3.2
Effects of control level compensation on tactical level
tasks to a braking lead vehicle
In reaction to the braking lead vehicle, participants drove at significantly lower
speeds on the directly intersections following the unexpected event than they
did on the same intersections in the standard drive (p<.0018). Figure 3 shows
the average speed on the intersection area (100 meters before until end of
intersection) measured over 7 intersections in drive 2 (braking lead vehicle) and
drive 3 (standard drive), and the interaction effects between drive and
intersection (p<.001). It can be seen that overall speed for these intersections is
significantly lower in the drive with the braking lead vehicle; the reaction to the
braking lead vehicle (intersection 5 in drive 2) is also clear.
Fig.3. Average driving speed after braking lead car and in reference condition.
Left: average speed in drive 2 and drive 3; Right: interaction effect between
intersection number and drive.
Lead car brakes at intersection 5 in drive 2
138
Modelling of drivers’ behaviour for ITS design
Furthermore, participants increased their headway significantly during the
second drive (p<.001), compared to the reference drives.
3.3
Effects of level of unexpectedness and additional
workload
The level of unexpectedness (L.O.U.) has a significant effect on the percentage
of time participants drove with the minimum headway (p<.038): when the lead
car would brake harder, the participants would drive less time with the minimum
headway (see Figure 4).
Fig.4. Effect of level of unexpectedness (lou) on % of time driven with minimum
headway.
1 = least unexpected, 3 = most unexpected
There was no significant influence found on driving speed of the level of
unexpectedness.
The additional task increased PDT reaction time (p<.001) and number of
missed PDT signals (p<.001), and this increased workload influenced the effect
that was seen on driving speed: with the additional task, the speed decrease in
the second drive was significantly less than without the additional task (p<.027).
A full description of our experimental results will be published in [7].
4 Conclusions
It can be seen from our results that tactical level tasks are influenced by
operational level compensation tasks in case of an unexpected event, and that
this effect fades away after a certain amount of time. A significant effect of the
level of unexpectedness on this influence between levels of the driving task was
seen, and additional workload seems to change the relation between level s of
the driving task as well. A full description of conclusions regarding the structure
of the driving task and our conceptual model of intersection driving behaviour
will be published in [7].
139
Human Centred Design for Intelligent Transport Systems
5 Acknowledgments
This research has been conducted within the research centre Applications of
Integrated Driver Assistance (AIDA) from the University of Twente and TNO the
Netherlands, and as part of the Intelligent Vehicles project of the Dutch
TRANSUMO program. TRANSUMO (TRANsition SUstainable MObility) is a
Dutch platform for companies, governments and knowledge institutes that
cooperate in the development of knowledge with regard to sustainable mobility.
We would also like to thank Jeroen Hogema and Marika Hoedemaeker (TNO
Defence, Security and Safety, BU Human Factors) for their help with the
experimental design. Furthermore, we are grateful to TNO for their help with the
driving simulator, and ST Software for their help with the implementation of the
experiment.
6 References
[1]
[2]
[3]
[4]
[5]
[6]
[7]
CBS: statistiek van het Nederlandse motorvoertuigenpark, Kennisbank
SWOV,
http://www.swov.nl/nl/research/kennisbank/inhoud/00_trend/05_mobiliteit/o
ntwikkelingen_in_het_voertuigpark.htm, accessed September 2007.
Schaap, T. W. and B. van Arem. An evaluation of behaviour simulation
modelling tools for urban intersection driving. TRAIL Congress, Rotterdam,
the Netherlands, November 2006.
Michon, J.A. (1971) Psychonomie onderweg. Inaugural speech at
University of Groningen. November 2, 1971.
Alexander, G.J. and H. Lunenfeld. Driver Expectancy in Highway Design
and Traffic Operations. Publication No. FHWA-TO-86-1. Federal Highway
Administration, Washington, D.C., 1986.
Van Winsum, W., M. Martens, & L. Herland. The effects of speech versus
tactile driver support messages on workload, driver behaviour and user
acceptance, Report TM-01-D009, TNO Human Factors Research Institute,
Soesterberg, The Netherlands, 1999.
Van Winsum, W. & M. Hoedemaeker. A road test of a prototype satellite
system for in-vehicle menu control, Report TM-00-C003, TNO Human
Factors Research Institute, Soesterberg, The Netherlands, 2000.
Schaap, T.W., A.R.A. van der Horst & B. van Arem. Studying behavioural
adaptation in urban traffic, Report University of Twente. Enschede, The
Netherlands [to appear].
140
Modelling of drivers’ behaviour for ITS design
SUBJECTIVE STRAIN ESTIMATION DEPENDING
ON DRIVING MANOEUVRES AND TRAFFIC
SITUATION
Caroline Schießl (DLR, Germany)
caroline.schiessl@dlr.de
ABSTRACT: An inadequate level of workload while driving is leading to
increased errors and accidents. Therefore, there is a need for levelling the
driver’s workload through some assistance system. In addition to their specific
functionalities (e.g. coordinate, prioritise information), these systems aim to
optimize the driver’s workload. Therefore, an assessment of the workload is
required. As workload depends on external stress factors which can change
dynamically, e.g. driving manoeuvres or environmental conditions, the workload
estimation needs to be as dynamic and continuous rather then discrete. In this
paper, the effects of traffic density and changes in the demands within a
complex manoeuvre are estimated using a new developed method for a
continuous subjective rating of the driver’s workload. The results demonstrate
that the variation of the stress factors moderate driver’s strain. By integrating
these findings with former results a qualitative and quantitative model of stress
and strain is introduced.
1 Introduction
1.1
Workload measurement
environment context
within
the
driver-vehicle-
Workload is a very complex concept of interrelations between e.g. external task
demands, internal resources, processing capacities and performance
capabilities. These aspects are often indicated with the same term “workload”
[1]. The approach presented here is based on a concept of stress and strain.
According to DIN EN ISO 10075-1 [2] psychological stress is defined as “the
total assessable influence impinging upon a human being from external sources
and affecting it mentally”, whereas strain is defined as “the immediate effect of
mental stress on the individual (not the long-term effect) depending on his / her
individual habitual and actual preconditions, including individual coping styles.”
While driving stress factors arise from different sources: First of all, the driving
task poses differing demands on the driver. Follow a road requires primarily
lateral control whereas following a preceding car additionally requires
longitudinal control actions. The second source for increased stress is the
environment modifying the requirements of the driving task as well as acting
directly on the driver. For example follow a road during fog should likely be
more difficult and strenuous than under the condition of high visibility because
the availability of relevant information is restricted. On the other hand, e.g. high
temperature influences the driver directly by making him / her tired. A third
category of stress sources are additional secondary tasks like operating a
141
Human Centred Design for Intelligent Transport Systems
navigation system, using a mobile phone [3, 4] or communicating with
passengers. The actual level of driver’s strain evoked by the stress depends on
driver characteristics such as abilities, skills and his state. Therefore, the same
stress level does not necessarily result in the same strain for different drivers.
Due to the vehicle motion within the driving environment and the interaction with
other road users none of the three stress sources acts statically on the driver.
They can be characterised by different distinctive changing dynamics which
again demands a comparable high dynamic description of the possible resulting
strain.
In general, strain cannot be measured directly but indirectly and is normally
assessed multimodal [1, 5] by subjective, physiological and performance
indicators. Behavioural indicators, e.g. steering wheel reversals or the standard
deviation of the lateral position indicate individual coping strategies in terms of
action control. Physiological measures such as the heart rate describe the
driver’s state by indicating his activation or arousal [1, 6]. Both indicators
dynamically assess changes in strain but have problems in regard to are partly
problematic with regard of sensitivity and specificity. This makes interpretation
of these indicators sometimes very difficult. Indicators e.g. have to be differently
interpreted according the test design (primary versus secondary task) but also
with regard to different sources (changing of the driver state versus changing of
the task demands) [7]. Besides, both types of indicators require special sensors
in the vehicle which are not available in current series-production vehicles but
would be mandatory for a wider application of strain adaptive systems not only
in the field of research. The third group of subjective self report measurements
like NASA-TLX or Instantaneous Self Assessment Method [8] don’t need any
specific sensors for assessment and in general have a higher sensitivity than
the indicators mentioned above. However, these indicators are usually collected
at discrete points in time like at the end of a test drive or at certain spatial- or
time-triggered situations during the trip to analyse differences between
systematically varied independent variables, system configurations, vehicles or
drivers. This event-triggered and discrete assessment of the data is not suitable
for measuring the effects of dynamically changing stress factors over time. In
recent own real vehicle driving studies [9] it was shown that driver strain varies
not only between different consecutive driving manoeuvres but even within one
manoeuvre. Therefore strain should be analysed in regard to structural changes
in place of level differences between certain time- or event-sections only by
means of a likewise dynamic and continuous measurement. The second reason
for a continuous measurement of both stress and strain is the final objective of
this research: the definition and implementation of an online working model of
driver stress and strain to manage the human machine interaction. These
systems require continuous access to the workload level and its changes in
order to provide an optimal support for the driver.
1.2
Estimation of strain by continuous subjective rating
measurement
To benefit from the higher sensitivity of subjective self-report measures and to
avoid the disadvantages of the rather discrete time- or spatial-triggered
methods, a continuous subjective rating method was developed and tested in
142
Modelling of drivers’ behaviour for ITS design
different former experimental studies. Within a real car driving study subjective
strain was measured subsequent to the test drive through a video analysis
where 16 participants rated their experienced strain during the drive using a 15point rating scale (see Figure 1). The rating scale consist of five verbal main
categories (very little strenuous, little strenuous, moderate, strenuous, very
strenuous) with 3 subcategories each. Participants were not informed about the
underlying stress factors varied by different driving manoeuvres and
environmental factors, to not influence their ratings. They were instructed to
give a new rating whenever they perceived a change of their subjective strain
and their actual rating was displayed within the video.
little strenuous
very little strenuous
1
2
3
4
5
moderate
6
7
8
strenuous
9
10
11
very strenuous
12
13
14
15
Fig.1. 15-point rating scale of the rating method
After mapping the continuous ratings to the different driving manoeuvres and
environmental factors the results of the study demonstrated that these stress
factors were associated with significantly different Perception of strain [9].
Different driving manoeuvres with different requirements were identified as
stress factors leading to different strain levels. As an example, approaching and
following a preceding car was shown to cause more strain than just following a
road as these manoeuvres pose, additional to lateral control, also longitudinal
regulation demands on the driver. However, situational characteristics were
shown to modify the effects. For example strain increases only during
approaching and following a preceding car in situations where the driver had the
possibility to overtake and prepared this manoeuvre. In addition the increase
was larger with oncoming traffic (rural road) and difficult road characteristics,
such as narrow lanes and curves, which posed extra demands on the lateral
regulation and velocity adaptation. In road sections where overtaking was not
allowed, these manoeuvres did not differ in strain. In these situations the
existence of the preceding car seemed to rather support the driver in
longitudinal regulation than posing additional demands on the driver. The
relationship between the described stress factors and subjective strain was
validated within the study by analysing the physiological indicator heart rate
variability as well as the performance indicator steering wheel reversal rate
which both resulted in comparable results.
As continuous subjective strain was measured post-hoc subsequent the test
drives, it was not clear whether memory effects or the restricted availability of
information within the video display influenced the results. With the intention to
estimate these effects and to test if it is possible to rate subjective strain
continuously while driving (online) a second study was conducted [10]. In order
to vary different stress factors within the DLR Virtual Reality Laboratory (VR
Lab), different road sections were constructed and surrounding traffic was
implemented so that test drivers conducted different driving manoeuvres on
different road sections for several times repeatedly. Resulting strain was
measured multimodal and continuously by subjective (by the abovementioned
rating method) as well as performance indicators. To analyse possible
interferences between driving and rating participants were divided into different
groups, two experimental groups that rated first simultaneously while driving
and secondly after the test drive by means of a playback-function. A control
143
Human Centred Design for Intelligent Transport Systems
group rated exclusively offline. Influences of the rating time (online vs. offline)
on subjective strain were analysed by comparing the ratings of the experimental
groups. To control the effect of different rating modes, a manual and a verbal
rating input have been compared additionally. It could be shown that a
continuous assessment of the subjective strain is not differing between the
online and offline approach. The analyses of the driving performance indicators
(e.g. average speed, standard deviation of lateral position) showed no statistical
significant interferences between rating and driving although the online rating
groups tended to decrease their driving speed with the additional rating task.
Furthermore no significant main effect of rating time (online versus offline) and
the rating mode (manual versus verbal input) on rating behaviour was found.
However, there were significant interactions between these factors and very
strenuous situations (e.g. the situation overtaking on the rural road). The control
group which exclusively rated offline underestimated strain as well as the online
rating group that rated by means of the manual input. Difficulties by only posthoc rating may be due to difficulties in the perception of distances and velocities
within the playback function of the VR Lab. The manual online rating seemed to
be somewhat difficult in situations where participants had to focus on motor
actions during the overtaking for example. On the other side these participants
adjusted their underestimated ratings afterwards during offline rating, which led
to a significant interaction in the comparison of online and offline rating
behaviour of the experimental group. Therefore it can be subsumed that if the
driving situation is not characterised by extremely high demands for the driver,
online rating is applicable and preferred to exclusively offline rating. Whereas in
very high demanding situations participants should have the possibility to adjust
their online ratings by an additional offline rating. The influence of the varied
stress factors on strain and its intercorrelations found in the first study were
validated within the second study.
To summarize the results of both former studies it was demonstrated that the
developed method to assess subjective strain continuously is an alternative to
time- or event-triggered self report measures. By using this method, dynamic
subjective strain information which is related to dynamically changing stress
factors can be collected. Concerning the rating method (online or offline), it can
be suggested to chose the method dependent on the experimental setup, i.e.
whether it is a real driving tests or a simulator study, as both approaches result
in comparable results.
1.3
Relation between stress and strain taking into account
different levels of dynamics
Within the studies mentioned above the examined stress factors can be
described as being of relatively low dynamics as within the real driving study
e.g. participants passed within one hour of test drive only through seven
different road sections. On the other hand, mapping subjective strain ratings on
different driving manoeuvres also taking into account the manoeuvre sequence
means estimating strain on a higher, more moderate dynamic level as the
participants conducted several manoeuvres within the different road sections.
For example within the abovementioned real driving study 16 participants
conducted within a one hour test drive a total of 3452 driving manoeuvres, 1130
144
Modelling of drivers’ behaviour for ITS design
on the motorway, 32 on the motorway exit, 701 on the city road and 1589 on the
rural road. Additionally there were indices that strain does not only differ among
the manoeuvres but also even within a manoeuvre where the dynamic of
changes is even higher and can be described as short-term changes. Within the
following study higher dynamics of stress factors changes will be examined.
The other main issue of the research is validating the method for different
dynamically stress factor changes, testing the sensitivity of the method and as a
prospective objective establishing a qualitative and quantitative relationship
between the measures of external stress factors and driver strain.
2 Method and test design
The main objective of the simulator study was to see whether even short time
changes of external stress factors could be represented by means of the
continuous rating method. As an environmental stress factor, traffic density was
varied. By this the influence of different traffic densities on driver strain while
driving on the motorway in general was measured. As former studies indicated
that strain even differed within a complex manoeuvre the second aim was to
assess the influence of different requirements within the complex manoeuvre
“entering the motorway” on the driver’s strain. To systematically distinguish
between different demands “entering the motorway” was divided into four
subtasks: First the orientation phase (at the beginning of the slip road), second
the planning phase were participants begin to search for appropriate gaps,
estimate velocities and plan their lane change during driving parallel to the other
vehicles on the acceleration lane, third the active lane change phase
characterised by the actual execution of the lane change and the last phase of
driving on.
13 subjects (8 male and 5 female) between 24 and 35 years old (mean = 28)
participated in the simulator study. The DLR simulator SimCar is characterised
by a dynamic motion system and a high quality projection system that provides
the image of the surrounding virtual environment and traffic. A wide field of view
(240° × 40°) combined with a high resolution of approx. 9200 × 1280 pixels
provides a detailed visual representation. For an overall realistic impression a
complete vehicle has been integrated into the simulator. Driver behaviour (e.g.
acceleration, steering) and the resulting vehicle dynamics (e.g. lateral
acceleration, velocity) are recorded as well as information from the
environmental model (e.g. preceding cars, road characteristics). The test track
consists of two different parts: The first part is characterised by a set of two
driving situations on the motorway with increasing and decreasing local traffic
density in order to analyse the general influence of changing traffic on driver
strain. The second part consisted of six consecutive manoeuvres “entering on
the motorway” with fixed different local traffic densities. The order of increasing
and decreasing traffic within the first part as well as the sequence of density
levels within the second part was balanced between and within the subjects in
order to control time- and cross-over effects. A total of about 1 hour of driving
was realised within the simulator study.
Local traffic density was implemented by the traffic model introducing different
numbers of vehicles. Local traffic density was defined by the coverage level of
145
Human Centred Design for Intelligent Transport Systems
vehicles in front and besides the own vehicle and was then classified into 10
stages from 0 (no traffic) to 1 (traffic congestion) [11]. In order to allow analyses
with repeated measures and to ensure equal traffic conditions for all participants
during the second part of the test drive (the six consecutive manoeuvres
“entering the motorway”) the coverage level was classified and applied by six
conditions of different levels of traffic density. The six traffic densities were
defined according the distinction of the different LOS-levels LOS1 (no traffic) to
LOS6 (traffic congestion) [12].
Driver strain was measured continuously and multimodal by subjective
(assessed by means of the online rating method with manual input),
physiological (e.g. heart rate) and performance indicators (e.g. parameters of
lateral control). Regarding the continuous rating method the participants were
instructed to change their subjective strain ratings by pressing a steering wheel
button whenever they experience a change in strain. Their actual rating was
displayed in a head-up display on the 15-point rating scale and the verbal
description of the main category (e.g. little strenuous, very strenuous). In order
to ensure safe driving with simultaneous subjective rating and to counteract
possible simulator sickness, the participants have been trained intensely to get
used to the simulator and the rating task.
3 Results
In order to analyse the overall influence of dynamically changing traffic densities
on subjective and physiological strain, these indicators were mapped to the
different coverage levels while driving on the motorway and the 6 LOS levels on
the slip road by computing mean values for each level of the factors.
Concerning the different demands of the manoeuvre “entering the motorway”,
the indicators were mapped likewise and mean values dependant from both
factors, the manoeuvre phase and the LOS-level were computed for each
participant (e.g. mean subjective strain during the orientation phase with high
traffic density LOS6).
Although systematic control of the environment is an advantage of simulator
studies, sufficient high coverage levels (> 0.7) of local traffic density on the
whole motorway were not realised for all test drivers. This is probably due to the
fact that the local traffic density strongly depends on the interaction between
vehicles and therefore on the individual driving behaviour of the test drivers and
the individual interaction with surrounding traffic respectively. As a consequence
the number of analysed coverage levels is only 7 (0 to 0.6) and only 6 out of the
13 subjects can be included in the analyses concerning the influence of traffic in
strain during the whole drive on the motorway. Therefore the results have to be
interpreted as trends. The heart rate values have been intra-individually zstandardised across the whole data to control different inter-individual
physiological initial levels. Subjective ratings have not been standardised as the
inter-individual variability was marginal by reason of the extensive training
phase.
In order to describe the overall influence of dynamically changing traffic density
on strain two separate two-way 2×7 Design ANOVAS with repeated measures
(sequence of increasing / decreasing traffic density, 7 levels of traffic density
146
Modelling of drivers’ behaviour for ITS design
coverage, α = 5%) were performed for the subjective ratings and the heart rate.
Concerning the subjective rating the effect of “traffic density” was significant
(F6,30 = 8.720, p < .001) whereas the second main effect “sequence” (F1,6 =.335,
p = .588) and the interaction between “traffic*sequence” (F6,30 =.891, p = .514)
were both not significant. The pairwise comparisons demonstrated that
subjective strain increased with raising traffic density up to a medium coverage
level of 0.3 but then remained on this level without increasing any higher. The
equivalent analysis was computed for the intra-individual z-standardised heart
rate (mean). The results tend to demonstrate comparable effects (traffic density:
F6,30 = 2.304, p = .060, sequence: F1,5 =.424, p = .544, traffic*sequence: F6,30
=1.048, p = .415). Mean heart rate also increased significantly with raising traffic
density up to a medium level but in contrast to the subjective rating decreased
again with even higher traffic, whereas the decrease was not statistical
significant. In Figure 2 both strain indicators are displayed. 1
subjective and physiological strain
physiological
strain_zvalue [0-1]
subjective
0
0.1
0.2
0.3
0.4
0.5
traffic density [coverage level 0 - 1]
0.6
Fig.2. Strain dependent from traffic density (n = 6)
Besides this overall effect, the influence of different traffic densities on strain
during the complex manoeuvre “entering the motorway” is described within the
following analyses. The 6 different traffic densities on the slip road and the
different demands of the manoeuvre can be characterised as of higher
dynamics as the increasing and decreasing traffic density during the drive on
the motorway. The analysed factors are therefore traffic density by means of the
LOS levels and the 4 different phases of the manoeuvre “entering the
motorway”. For both strain indicators two separate two-way 6×4 Design
ANOVAS with repeated measures (6 LOS levels, 4 phases, α = 5%) were
performed. Due to technical problems not all traffic densities have been realised
completely in the simulation and therefore only 11 out of the 13 subjects can be
included in the analyses. Regarding the influence of the different LOS levels the
main effects are significant in both analyses (mean subjective rating: F5,50
=3.484, p = .009, mean heart rate: F5,50 =4.947, p = .001). Subjective strain
increases with increasing traffic, whereas only the lower stages differ
significantly from the higher ones with subjective strain reaching a plateau.
Physiological strain as well increases up to a medium traffic density but then
1
By reason of a better demonstration z-values of both indicators are displayed
within the figure. The analyses of subjective z-values result in comparable
findings as the absolute ratings.
147
Human Centred Design for Intelligent Transport Systems
decreases again. The difference between the lowest level and the others is the
only significant result. The factor “phase” has a significant influence on strain for
the subjective indicator (F3,30 =13.406, p < .001) whereas only a tendency is
found for the z-standardised heart rate (F3,30 =2.786, p = .058). Both indicators
increase up to phase 3 follow by a decrease so that the highest strain is related
to the active lane change in both indicators (see Figure 3).
subjective and physiological strain
physiological
strain_zvalue [0 - 1]
subjective
Phase1
Phase2
Phase3
Phase4
Fig.3. Strain dependent from the manoeuvre phases (n = 11)
Furthermore, a significant interaction between LOS level and manoeuvre phase
for both indicators (mean subjective rating: F15,150 =4.338, p = .000, mean heart
rate: F15,150 =2.665, p = .001) can be found.
Figure 4 demonstrates the interaction between the both factors. Regarding the
subjective indicator strain increases during low traffic densities (LOS2 and
LOS3) in phase 2 (planning phase) and decreases during the actual lane
change and the continuation of the manoeuvre. From LOS3 strain remains on a
high level and increases even more with increasing traffic respectively. In phase
4 strain is comparable to during the actual lane change except during the
highest density level. Within the highest level (LOS6) strain decreases within
phase 4 which is probably due to the very low velocities within the congestion.
The interaction concerning the physiological indicator demonstrates a
somewhat different relation. At high density levels a comparable increase in
strain up to the lane change manoeuvre can be described, whereas at low
levels strain first decreases in phase 2 and then got its maximum in phase 3.
148
Modelling of drivers’ behaviour for ITS design
subjective [abs.-value, mean], LOS and phase
subj_m [1 - 15]
subj_m [1 - 15]
LOS1
9
8
7
6
5
4
9
8
7
6
5
4
LOS2
LOS3
D
$
D
$
D
$
$
D
D
$
D
$
$
D
$
D
D
$
$
D
D
$
$
D
LOS4
LOS5
LOS6
D
$
D
$
D
$
D
$
D
$
D
$
D
$
D
$
D
$
D
$
D
$
D
$
1
2
3
phase
4
1
2
3
phase
4
1
2
3
phase
4
zhr_m [bpm]
zhr_m [bpm]
heart rate (z-value, mean), LOS und phase
LOS1
1.0
LOS2
0.5
$
D
$
D
0.0
LOS3
$
D
$
D
$
D
D
$
D
$
D
$
D
$
$
D
$
D
-0.5
$
D
LOS4
1.0
LOS5
LOS6
$
D
$
D
0.5
$
D
$
D
$
D
$
D
0.0
D
$
$
D
$
D
$
D
$
D
$
D
-0.5
1
2
3
phase
4
1
2
3
phase
4
1
2
3
phase
4
Fig.4. Interaction between LOS level and manoeuvre phase (n = 11)
4 Discussion and perspective
The results of previous studies demonstrated that the developed approach of
continuous subjective strain measurement is applicable to assess the effect of
different stress factors (manoeuvre and situation) on driver strain in different
experimental setups such as real driving and simulation as well online as offline.
Furthermore it was demonstrated that the relation between the measurement of
varying external factors and the resulting driver strain, taking into consideration
different levels of dynamics can also be represented by the method. The
influence of short-term change of local traffic density is describable as well as
the influence of various demands within the complex manoeuvre “entering the
motorway”. Concerning the general influence of traffic density on driver strain
the analyses of the subjective and the physiological indicators show only partly
comparable results: During high traffic densities subjective strain remains on a
plateau whereas physiological strain decreases. This may be due to different
sensitivities of the mean heart rate regarding physical and mental load. Mean
heart rate is seen as sensitive to physical load but less sensitive to mental load.
149
Human Centred Design for Intelligent Transport Systems
It can be argued that mental load is higher than physical load in high traffic
density situations due to a restricted scope of action or behaviour within the
congestion. As a result the indicator of mean heart rate might be less sensitive
in high traffic density situations than in low traffic density situations. In order to
validate this interpretation other physiological indicators being more sensitive to
mental load as e.g. the heart rate variability, should be analysed in further
studies. Besides it can be stated that during high traffic densities strain
increases until the actual lane change, whereas strain reached its maximum
already during the planning phase in low densities, where the driver has to
estimate velocities and distances to other vehicles and decide the lane change.
The significant interaction demonstrates the importance to describe relations
among different dynamically changing stress factors and resulting subjective
driver strain more precisely. By means of the continuous subjective rating
method more detailed information about the influence of factors on strain can be
gathered than by a subjective questionnaire after the whole test drive which in
turn results in a better understanding of the analysed aspect. However the
dynamics should be analysed in more detail in further studies by examining the
changing sequences. To reach the general aim of the research to model stress
and strain within the driver-vehicle-environment context the relative influence of
different stress factors with different changing dynamics on strain can be
described and modelled by integrating the results of the studies. The final and
prospective aim of the research is then the indirect estimation of the related
strain via the underlying model of stress and strain by directly measuring the
involved stress factors via CAN-Bus (e.g. manoeuvre recognition) or laser
scanner (e.g. traffic density) in order to provide this information to possible
adaptive assistance or information systems.
5 References
[1]
[2]
[3]
[4]
[5]
[6]
[7]
De Waard, D.: The Measurement of Drivers' Mental Workload (Traffic
Research Center VSC, 1996)
DIN EN ISO 10075-1: Ergonomische Grundlagen bezüglich psychischer
Arbeitsbelastung, 2000
Alm, H., Nilsson, L.: The effects of a mobile telephone task on driver
behaviour in a car following situation, Accident Analysis & Prevention,
1995, 27, (5), pp. 707-15
Becker, S., Brockmann, M., Bruckmayr, E., Hofmann, O., Krause, R.,
Mertens, A., Nin, R., Sonntag, J.: Telefonieren am Steuer
(Wirtschaftsverlag NW, 1995)
Wickens, C.D., Hollands, J.: Engineering psychology and human
performance (Longman, 1999)
Hicks, T. G., Wierwille, W. W.: Comparison of Five Mental Workload
Assessment Procedures in a Moving-Base Driving Simulator, Human
Factors, 1979, 21, pp. 129-143
MacDonald, W.A., Hoffman, E.R.: Review of relationships between
steering wheel reversal rate and driving task demand, Human Factors,
1980, 22, pp. 733-739
150
Modelling of drivers’ behaviour for ITS design
[8]
Girard, J.M., Wilczyk, M., Barloy, Y., Simon, P., Popieul, J.C.: Towards an
On-line Assessment of Subjective Driver Workload. Driving Simulator
Conference, Orlando, USA, 2005, pp. 382-391
[9] Schießl, C., Vollrath, M., Altmüller, T., Dambier, M., Kornblum, C.: What is
strenuous? Driving itself or the driving situation? 13th World Congress on
Intelligent Transportation and Systems and Services, London, Great
Britain, October 2006
[10] Schießl, C.: Dynamic access to the actual driver state. 6 th European
Congress on Intelligent Transportation Systems and Services, Aalborg,
Denmark, Juni 2007
[11] Knake-Langhorst, S., Schießl, C.: Local traffic condition - improvement of a
vehicle based measurement approach. 6th European Congress of
Intelligent Transportation Systems and Services, Aalborg, Denmark, June
2007
[12] Special report 209: Highway capacity manual. Washington, D.C.:
Transportation Research Board, 1985
151
Modelling of drivers’ behaviour for ITS design
IDENTIFYING USER STRATEGIES FOR
INTERACTION WITH IN–VEHICLE INFORMATION
SYSTEMS WHILE DRIVING IN A SIMULATOR
Nadja Rauch, Barbara Gradenegger, Hans-Peter Krüger
(IZVW, Wuerzburg University, Germany)
rauch@psychologie.uni-wuerzburg.de
ABSTRACT: The study analyzes the efficiency of different strategies for
interaction with in-vehicle information systems. N=24 drivers completed a test
course in a motion-base driving simulator containing different critical situations.
At predetermined points of the route an additional menu navigation task was
offered to the driver. The driver could decide whether the actual situation was
suitable to execute a task and when to interrupt it. The results show that drivers
are able to adapt their secondary task behaviour to the situational demands.
The anticipation of potential conflicts could be shown both in secondary task
behaviour (complete task rejection or task delay in critical situations) as well as
in driving behaviour (e.g. lower approaching speed in front of demanding
situations). These strategies were successful to maintain driving safety.
Adequate situational assessment prior to the start of the task and adequate
monitoring of situational development during secondary task execution are
identified as relevant processes for Situation Awareness in this context.
1 Introduction
Extended research has been done on the effects of dealing with in-vehicle
information systems (IVIS) while driving. The overall result is that performance
of an additional secondary task clearly reduces driving performance and safety.
Typical effects are a decrease in lateral control (e.g. [1]) or delayed reaction
times to sudden events (e.g. [2]). On the other hand also compensation
strategies can be monitored, e.g. reduction of speed ([3] and [4]), an increase of
safety margins [5] or fewer lane changes [6]. An often neglected fact is that
drivers are also able to compensate additional workload by specific interaction
strategies dealing with the secondary task. Like McCartt et al. [7] argue: “phone
and driving tasks are paced by experimenters, but in the real world drivers
decide when and where to use their phones and may adapt their phone use to
varying traffic conditions” (p.92). This freedom of decision is often not given in
typical experimental studies analyzing the potential risk of dealing with
secondary tasks while driving. Some results from telephone interviews [8] and
video observations [9] hint to specific compensation strategies, for example
while using a cell phone. Some drivers do not use the phone in a moving
vehicle at all, some use it only while they are waiting at a red traffic light, some
even stop the car. While calling, drivers tend to drive more slowly, avoid lane
changes, choose sections with lower traffic density for calling someone or ask a
passenger to do the call. In demanding traffic situations drivers interrupt the
conversation, or the remote caller is informed about the environmental
153
Human Centred Design for Intelligent Transport Systems
conditions to adapt the conversation. While performing a secondary task,
drivers try to divide the tasks into smaller chunks und look back to the road in
adequate intervals [10].
2 Objectives
Compensation strategies in the interaction with IVIS while driving can be
executed on the level of driving task as well as on the level of secondary task.
The premise for those compensation strategies is to have Situation Awareness
of the driving task. This concept was originally developed in aviation but has
become more and more important also in the domain of driving. The most
famous definition is the one by Endsley who defines Situation Awareness as
“the perception of the elements in the environment within a volume of time and
space, the comprehension of their meaning and the projection of their status in
the near future” [11; p. 792]. It is argued that a cognitive representation of the
current and the future situation is formed on the base of knowledge in long-term
memory which guides attention to the relevant cues in the environment. Due to
the dynamics of the driving situation, this mental representation has to be
continuously updated. According to a model of Adams, Tenney and Pew [12]
this update is understood as a cyclical process, including an active search in the
environment for information which may prove or disprove the activated schema
and lead to modifications.
In the context of interactions with IVIS we postulate two different processes of
Situation Awareness necessary for a situationally adaptive use of secondary
tasks while driving:
1. An adequate assessment of the situation prior to the start of a secondary
task for proper decisions if a secondary task can be performed in the
current situation at all. In highly demanding traffic situations the driver
should ignore an additional task or should delay the beginning of the task.
2. An adequate monitoring of situational development during secondary task
execution in order to permanently compare the expected situational
development with the actual one. In case of observed differences, the
secondary task has to be interrupted.
We argue that the control process during secondary task execution is restricted
to a subjectively relevant part of the environment. Therefore, the anticipative
process of situational assessment prior to task execution is crucial for the
proper control of situational development. It influences which of the cues in the
environment will be monitored during secondary task execution.
In this paper we try to identify the two postulated processes of Situation
Awareness in the driving task as well as in the interaction with secondary tasks
while driving. Therefore, we examine anticipative compensation strategies in the
decision process just before the beginning of a task as well as more reactive
compensation strategies during secondary task execution and analyze their
effects on driving safety.
154
Modelling of drivers’ behaviour for ITS design
3 Method
3.1
Driving task
To examine different compensation strategies in interactions with IVIS, a study
in a driving simulator with motion system of the WIVW (Wuerzburg Institute for
Traffic Sciences GmbH, for more information see www.wivw.de) was
conducted. The 1.25 h test course consisted of sections with rural roads and
urban areas. Within this course, eight specific situations, including a potential
conflict, were realized (e.g. a vehicle parking out from a parking zone in front of
the subject’s vehicle, a pedestrian crossing the road just in front of the subject’s
vehicle, a broken-down car on a rural road, curvy sections etc.). Each of these
situations was realized three times varying in the salience of environmental
cues pointing to the specific conflict (good, medium, hard to predict; not further
discussed in this paper). For all these situations, it was characteristic that the
conflict could possibly be anticipated by the driver. Figure 1 shows the driving
simulator of the WIVW (left) and a screenshot of a simulator scene of a typical
urban situation (right).
N=24 subjects participated in the study: one control-group with a baseline
driving condition (n=8 subjects) and one dual task group which performed an
additional menu navigation task while driving (n=16 subjects). The mean age of
subjects was 36 years (sd=10.1 years) with n=12 female and n=12 male
drivers. The two groups did not differ in relevant trait factors like driving
experience, age or gender. They were all well trained in simulator driving.
Fig.1. The driving simulator with motion system of the WIVW (left) and a
screenshot of a typical driving scene (right).
3.2
Menu navigation task
In the dual task condition, the driver had to perform a secondary task while
driving. The presentation of this task was the following: At predetermined points
of the route, the driver was offered the choice to perform an additional task. This
offer was given either just before a critical situation (e.g. just after a parked car
started to indicate but was still standing) or in a non-critical situation (on road
segments between the critical situations). The offer was signalled by a question
mark shown in the Head-Up Display on the front scene (see figure 2 left).
155
Human Centred Design for Intelligent Transport Systems
Fig.2. Screenshot of the HUD for offering a secondary task (left) and the menu
selection task (right).
The offer given, the driver had to decide within 3 seconds whether the situation
was suitable for the secondary task or not, according to the situational demands
(‘decision phase’). To start the secondary task, the driver had to pull a joystick
on the middle console of the vehicle to the right (see figure 2 right). Then a
“start display” was presented on a visual display located at a lower position on
the middle console of the vehicle (‘start phase’). To reject the offer, nothing
must be done. After 3 seconds the question mark disappeared and the driver
had to wait for the next opportunity to perform a task.
The task itself was a hierarchical menu navigation task simulating the
interaction with a typical in-vehicle information system. The task could be
interrupted at any point within the menu navigation. The driver was instructed to
navigate to a specific menu function (e.g. “control average fuel resumption”;
‘instruction phase’). To navigate within the menu, the joystick was used. After
reaching the correct function the driver confirmed the selection and the task was
finished (‘performance phase’). For limiting the time for task performance until
the next task was offered, menu navigation was stopped automatically after 15
seconds. The driver was instructed to interrupt the secondary task and return to
the driving task whenever the driving task required full attention. To motivate the
subjects to execute tasks at all, they received rewards for every completed task
but also penalties for severe driving errors.
4 Results
In a first step, compensation strategies in the secondary task (ST) were
analyzed by the following parameters of the menu task:
•
Mean number of rejected tasks [%]
•
Mean preparation time before starting menu navigation (start phase +
instruction phase), only for executed tasks [sec]
•
Mean execution time for menu navigation (limited to max. 15 sec), [sec]
•
Number of tasks with minimum one interruption (delay times between
two steps within the menu system > 3 sec), [%]
156
Modelling of drivers’ behaviour for ITS design
mean % of rejected tasks
90,0
7,0
non-critical
critical
6,0
mean preparation time [s]
100,0
80,0
70,0
60,0
50,0
40,0
30,0
20,0
non-critical
critical
5,0
4,0
3,0
2,0
1,0
10,0
0,0
0,0
urban
rural
urban
rural
Fig.3. Mean % of rejected tasks (left), mean preparation time (right), for urban
and rural, critical vs. non-critical situations.
The percentage of rejected tasks was generally higher in critical situations (43%
vs. 17%; see figure 3 left), especially on rural roads (in front of curves and the
broken-down vehicle; significant main effects “criticality” F[1;14]=74.737, p<.000
and “road type” F[1;14]=6.217, p=.026 and significant interaction;
F[1;14]=33.023; p<.000). Drivers seemed to anticipate these demanding
situations and therefore avoided any additional load of dual task performance. If
drivers decided to execute a task in critical situations (after all over 50% of
tasks) they try to await how the situation will develop (see figure 3 right). This
resulted in longer mean preparation times until the menu system was started in
front of critical situations (significant main effects “criticality“ F[1;14]=54.613,
p<.000 and „road type“ F[1;14]=8.399; p=.012 and significant interaction
F[1;14]=5.109; p=.042).
10,0
50
non-critical
critical
45
% taks with interruptions
mean performance time [s]
12,0
8,0
6,0
4,0
2,0
non-critical
critical
40
35
30
25
20
15
10
5
0,0
0
urban
rural
urban
rural
Fig.4. Mean performance time (left) and % tasks with interruptions (right), for
urban and rural, critical vs. non-critical situations.
The execution time for menu navigation was also longer in critical situations,
especially on rural roads (significant main effects “criticality“ F[1;13]=9.756;
p=.008 and „road type“ F[1;13]=27.214; p<.000 and significant interaction
F[1;13]=28.016; p<.000; see figure 4 left). This can be referred to a greater
157
Human Centred Design for Intelligent Transport Systems
number of necessary interruptions during ST execution in these situations. The
results for task interruption didn’t reach significance (“criticality“ F[1;14]=2.671,
p<.124, „road type“ F[1;14]=3.877; p=.069 and interaction F[1;14]=2.011;
p=.178).
Figure 5 shows the relevant driving parameters in the different phases of ST
execution, separated for straight and curvy roads. In both conditions, speed is
continuously reduced, beginning with the decision phase until the end of ST
execution. This implies that driving behaviour is first of all directed by the driving
task, but then adapted to the requirements of ST execution.
The parameters SDLP/second and steering wheel_sd/second (due to variable
phase durations standardized to sd/second) show an increase in steering
activity at the beginning of the secondary task which may be interpreted as an
attempt to get an optimum lane position, anticipating that during ST execution
lateral control will be worse. During the execution of the secondary task,
steering activity was then reduced which can be interpreted as a distraction
effect of the secondary task that lead to a disregard of lateral control in the
driving task. After the ST execution, often larger steering corrections could be
observed in order to adjust the accumulated lane position error during the ST
execution.
0,25
27,0
0,20
26,0
SDLP/sec [m]
mean velocity [m/s²]
28,0
25,0
24,0
23,0
22,0
0,15
0,10
0,05
21,0
20,0
straight
curve
after
instr
perform
start
after
decision
instr
straight
perform
start
decision
after
perform
instr
start
decision
after
perform
instr
start
decision
0,00
curve
steeringwheel_sd/sec [°]
0,12
0,10
0,08
0,06
0,04
0,02
straight
after
perform
instr
start
decision
after
perform
instr
start
decision
0,00
curve
Fig.5. Mean velocity, mean standard deviation of lateral position (SDLP) per
second and mean standard deviation of steering wheel activity per second in
different phases of the secondary task execution on straight and curvy rural
sections. Decision=decision phase, start=start phase, instr=instruction phase,
perform=task performance phase, after=phase 2.5 s after ST performance.
158
Modelling of drivers’ behaviour for ITS design
For analyzing the effects on driving safety, we defined performance strategies in
the interaction with IVIS in critical situations based on the following groups:
•
Rejection of a task (not performing the task at all)
•
Delayed beginning of the task (beginning of the task after a
critical point; defined separately for each of the situations;
e.g. the moment when the other vehicle is just parking out)
•
Execution of the task without any interruptions
•
Interruption of the secondary task, implicating a return to
the driving task after the secondary task was started
15
10
Baseline
5
interrupted
without
interruption
delayed
rejected
0
interrupted
25
20
without
interruption
30
delayed
35
50
45
40
35
30
25
20
15
10
5
0
rejected
40
% driving error
% of all offered tasks
45
performed
Fig.6. Percentage of different performance strategies for secondary task offers
in critical situations (left), percentage of situations with driving errors of all
critical situations; separately shown for the different performance strategies.
Figure 6 left gives the percentages of how often the different performance
strategies were selected related to all offered secondary tasks in the defined
critical situations. In most cases the task was rejected completely (43%). In 30%
the beginning of the task was delayed until the critical event had happened.
Further 14% of the tasks were performed without any interruptions straight
through the critical situation. In 13% of the tasks the task was interrupted after
the beginning of the task.
To prove the efficiency of the different strategies, driving errors were defined as
decelerations > 8 m/s², critical safety margins < 1 m or critical TTC < 1 sec.
Figure 6 right gives the results for the different strategies compared to the errors
in the baseline condition without secondary tasks. If the task was rejected or
delayed, there was no important increase in driving errors compared to the
baseline condition (baseline-rejected Chi²=0.019; p=.890; baseline-delayed
Chi²=1.765; p=.184). If the secondary task was performed straight through the
critical situation there was a visible increase in driving errors (baseline-without
interruptions: Chi²=3.596; p=.058). The highest percentage of driving errors
occurred if the task had to be interrupted after it was started (baselineinterrupted: Chi²=20.748; p<.000). Evidently, the interruption of the secondary
task was not successful to avoid driving errors.
159
Human Centred Design for Intelligent Transport Systems
5 Conclusions
Results show that drivers are able to adapt their secondary task behaviour to
the situational demands. The anticipation of potential conflicts can be shown
both in secondary task behaviour (task rejection or task delay) as well as in
driving behaviour (lower approaching speed or preparation of proper lane
positions in front of demanding situations like curves). These strategies seem to
be very successful in maintaining an adequate driving safety. We observed less
driving errors in critical situations when no secondary task was offered
(baseline), when the secondary task was rejected or when the beginning of the
task was delayed.
During the execution of a secondary task, drivers try to further adapt their
behaviour to the situational development. They further reduce speed, e.g. on
curvy sections and interrupt a task if the situation gets critical. In contrast to
anticipative strategies, the control of situational development after the beginning
of the task seems to be much more difficult. The results show that drivers who
didn’t’ use anticipative strategies and performed the menu task straight through
the critical situation, had a higher risk of driving errors, e.g. colliding with a
suddenly braking preceding vehicle.
Also an interruption of ST performance was not a successful strategy in this
study. In fact, the interruption of the task is an indicator of an adequate
monitoring of the situational development during ST execution. But we believe
that the success of this strategy is largely based on the time criticality of
situations. Most situations in this study were so time-critical that an interruption
of the menu navigation after starting the task could only be performed reactionbased but not anticipation-based. In less time critical situations a timely
interruption of the secondary task execution should be successful to avoid
negative effects on driving safety.
The still high number of executed tasks in critical situations indicates the
potential problems of the interaction with in-vehicle information systems while
driving. Drivers tend to perform secondary tasks also in situations where the
load of the driving task is already very high and then may have problems to
maintain driving safety. First of all, we explain this finding with an inadequate
assessment of the situation’s potential risk. In further analyses on the influence
of situation’s predictability in the study, varied by the availability of
environmental cues pointing to the specific conflict, it could be shown that the
less easy it was to predict the situation’s criticality the higher the number of
executed tasks was in this situation. This result indicates special problems for a
situationally adaptive use of in-vehicle tasks in real traffic situations with low
predictability. Another explanation would be that motivation for secondary task
may have been very high (drivers got rewards in the form of bonus points for
every completed task; the subject with the highest number of points at the end
received a gratification) despite the instruction that primary task should be
prioritized (they received a deduction of points for severe driving errors, e.g.
crossing the lane). For real traffic this indicates a special risk for tasks with high
relevance for the driver (e.g. important job-related phone calls). Furthermore,
drivers may have underestimated the associated risk of secondary tasks in
general. Results from a special questionnaire on the general attitudes of the
160
Modelling of drivers’ behaviour for ITS design
drivers towards the use of secondary tasks while driving indicate that drivers
who rated secondary tasks less distracting, rejected fewer tasks in the study.
In summary, the results show the importance of the two postulated processes of
Situation Awareness for situationally adaptive interactions with IVIS while
driving. The adequate situational assessment prior to ST execution is crucial to
avoid overload due to the additional load of a secondary task in already
demanding driving situations. Monitoring of situational development during ST
execution is also necessary but much more difficult for maintaining Situation
Awareness in interactions with IVIS.
The specific experimental setting, in which the driver decides whether and how
to interact with a secondary task, allows a realistic assessment of driver
interaction strategies with in-vehicle information systems. It permits analysis of
specific compensation strategies and of their effects on driving performance and
safety. Not only can possible distraction effects of an additional task be
explored, but also the conscious decisions for the neglect of the driving task and
their correlation to Situation Awareness.
6 Acklowledgements
The study was conducted within a project financed by the German Association
for Research on Automobile Technique and the Federal Highway Research
Institute.
7 References
[1]
[2]
[3]
[4]
[5]
[6]
[7]
Törnros, J.E.B. and Bolling, A.K.: ‘Mobile phone use-effects of handheld
and handsfree phones on driving performance’, Accident Analysis and
Prevention, 2005, 37, pp. 902-909
Strayer, D.L., and Drews, F.A.: ‘Profiles in driver distraction: Effects of cell
phone conversations on younger and older drivers’, Human Factors, 2004,
46, pp. 423--428
Horberry, T., Anderson, J., Regan, M.A., Triggs, T.J. and Brown, J.: ‘Driver
distraction: the effects of concurrent in-vehicle tasks, road environment
complexity and age on driving performance’, Accident Analysis and
Prevention, 2006, 38, pp. 185-191
Jamson, A.H. & Merat, N.: ‘Surrogate in-vehicle information systems and
driver behavior: effects of visual and cognitive load in simulated rural
driving’, Transportation Research Part F, 2005, 8, pp. 79-96
Ishida, T. and Matsuura, T.: ‘The effect of cellular phone use on driving
performance’, International Association of Traffic Safety Sciences (IATSS)
Research, 2001, 2S, pp. 6-14
Beede, K.E. and Kass, S.J.: ‘Engrossed in conversation: the impact of cell
phones on simulated driving performance’, Accident Analysis and
Prevention, 38, 2006, pp. 415-421
McCartt, A.T., Hellinga, L.A. and Braitman, K.A.: ‘Cell phones and driving:
review of research’, Traffic Injury Prevention, 2006, 7, pp. 89-106
161
Human Centred Design for Intelligent Transport Systems
Boyle, J.M. and Vanderwolf, P.: ‘Motor Vehicle Occupant Safety Survey:
Volume 4: Crash Injury and Emergency Medical Services Report (Report
No. DOT HS-809-857)’, National Highway Traffic Safety Administration,
Washington, DC, 2005
[9] Esbjörnsson, M. and Juhlin, O.: ‘Combining mobile phone conversations
and driving –Studying a mundane activity in its naturalistic setting’. Paper
presented at the ITS World Congress 2003, Madrid, Spain
[10] Salvucci, D. D., and Macuga, K. L.: ‘Predicting the effects of cellular-phone
dialling on driver performance’, Cognitive Systems Research, 2002, 3, pp.
95-102
[11] Endsley, M.R.: ‘Design and evaluation for situational awareness
enhancement’. Proceedings of the Human Factors Society 32nd Annual
Meeting. HFES, Santa Monica, 1988, pp. 97-101
[12] Adams, M.J., Tenney, Y.J. and Pew, R.W.: ‘Situation awareness and the
cognitive management of complex systems’, Human Factors, 1995, 37 (1),
pp. 85-104
[8]
162
Modelling of drivers’ behaviour for ITS design
PARAMETERS RELATED TO MODELLING
INTELLIGENT SPEED ADAPTATION SYSTEMS
WITH THE EMPLOYMENT OF A MICROSCOPIC
TRAFFIC MODEL
Ioanna Spyropoulou, Matthew G. Karlaftis
(National Technical University of Athens, Greece)
iospyrop@central.ntua.gr
ABSTRACT: This paper addresses the incorporation of intelligent speed
adaptation (ISA) systems into microscopic traffic models. The suggested model
is Gipps’ car-following model and the study involves the identification of the
appropriate model parameters that need to be modified as well as the
necessary steps towards the design of the modified model. Driver behaviour
under the use of three different functionalities of ISA, namely – informative,
warning and intervening – was investigated with the conduction of a driver
simulator experiment. Specific parameters that would describe the
implementation of the system were identified including parameters related to
system operation and impact on driver behaviour. Driver behaviour as
described through these parameters was also quantified to confirm the
necessity of their modification or introduction into the model.
1 Introduction
Intelligent transport systems (ITS) comprise a fast developing field and are
anticipated to contribute to a more sustainable road traffic network by improving
road safety, traffic conditions, environmental conditions and mobility [1]. A
prerequisite of their successful implementation is determining their anticipated
impact which can be achieved with a variety of tools including on-road and
simulator studies, laboratory experiments and simulations based on traffic/driver
models. In relation to the latter tool, traffic models need to be updated to include
the operation of intelligent transport systems.
In the recent years effort has been made into incorporating intelligent transport
systems into traffic models. This offers the possibility of enhancing the utility of
the models and the opportunities of intelligent transport system evaluation and
consequently of the assessment of traffic management and road safety
strategies. However, current applications mainly involve advanced traveller
information systems, rather than advanced driver assistance systems, and more
specifically the impact of information provision on driver route choice and
consequently on road network conditions. Such examples include the traffic
simulation programs AIMSUN [2], CONTRAM [3], PARAMICS [4] and VISUM
[5]. In a rather different application [6] simulated the differences in lane
changing and gap acceptance behaviour at congested road sections when
drivers are informed of traffic incidents downstream, using the traffic simulation
model SITRAS [7]. Intelligent speed adaptation has also been incorporated in
163
Human Centred Design for Intelligent Transport Systems
the microscopic traffic simulation model DRACULA [8]. [9]. This application
however concentrates on the operation of the system itself and its
representation employing model parameters rather than the representation of
the impact of this operation on driver behaviour.
This study discusses the specific aspects of the impact of intelligent speed
adaptation systems on driver behaviour related to driving speed (speed,
acceleration and deceleration), and the possible ways of incorporating them in a
traffic model. The impact of the systems on driver behaviour is elicited from the
analysis of simulator data and the model used to illustrate the proposed
approach is Gipps’ car-following model [10].
2 Intelligent speed adaptation simulator study
2.1
System operation
A study at the Transport Research Laboratory driver simulator to investigate
driver behaviour under the use of intelligent speed adaptation (ISA) systems
was conducted, and three different ISA functionalities – related to the human
machine interface (HMI) were investigated: informative, warning and
intervening.
A pictogram indicating the prevailing speed limit as well as its justification was
transmitted through an in-vehicle screen during the informative ISA application.
The operation of the warning ISA involved a tone repeated 3 times (0.5 sec on,
0.5sec off) the first time that the speed limit was exceeded and a single tone
(0.5 sec) every 8 seconds that the driver was continuously exceeding the limit
were transmitted. In addition, the auditory warnings were also supported by
visual information of the prevailing speed limit through the in-vehicle screen.
Last, the Intervening ISA did not allow the driver to adopt speeds higher than
the prevailing speed limit. If the system determined that the accelerator
pressure would cause the vehicle to exceed the limit, the pedal value was
reduced automatically. A smooth deceleration was also applied when the
vehicle was about to enter a speed limit zone with speed higher than the speed
limit. The image of the prevailing speed limit was presented via the in-vehicle
screen and a 0.5sec tone was transmitted, whenever the intervening
functionality was triggered.
2.2
Simulation drives
Each driver made four drives, one without a system which will be referred to as
“base condition” and one with each of the three ISA functionalities operating.
The simulated road environment involved a 2-lane single carriageway road and
consisted of rural road sections with 60mph speed limit, road sections through
villages with 30mph speed limit and residential areas with 20mph speed limit.
The vehicle was driving unimpeded by other vehicles (i.e. no vehicles were
placed in front of it), with the exception of three different types of incidents
which were simulated within the 60mph speed limit areas. The first involved an
accident blocking the lane indicated by two stopped vehicles one of which was
an ambulance. The driver had to overtake the stopped vehicles using the lane
164
Modelling of drivers’ behaviour for ITS design
of the opposite direction. The second incident involved inserting a slow-moving
vehicle in-front of the simulator vehicle, and the third involved the appearance of
a green dot on the vehicle windscreen at which the driver was told – prior to the
drive – to react by reducing vehicle speed to 10mph. The characteristics of the
drives in terms of order of incidents and systems varied both within and
between drivers, so as to eliminate possible order effects.
2.3
Participant characteristics
24 participants holding a driving license were recruited from the Transport
Research Laboratory database and the quota applied for their selection
involved driver age, as the objective was to have a balanced sample between
young/novice drivers and experienced but not elderly drivers. Following the
experiment 23 drive-sets fulfilled the requirements for data processing; and the
characteristics of these drivers appear in Table 1.
Table1. Participant Characteristics
Age
(years)
31.22
17.00
46.00
8.57
Average
Minimum
Maximum
St. Dev.
Driving experience
(years)
Annual mileage
(miles)
13.06
0.67
28.00
8.53
12909
2000
36000
9506
3 Translation of implementation of Isa into traffic model
parameters
3.1
Gipps’ car-following model
Gipps’ car-following model is discrete-time, continuous-space microscopic
model. The model was initially intended for simulation of free-flow traffic, and
hence used for simulating motorway roads, but is now being employed in
various traffic simulation programs including MULTSIM [11], SIGSIM [12],
AIMSUN, DRACULA and SITRAS. It is quite detailed, and there is a trade-off
with the computational time, which is long by comparison with other simpler
models. The variables that are calculated in each time-step are the vehicle’s
speed and through this its new position. The formula that is used to calculate
the “updated” speed (speed at time t + τ ) is:
u n (t + τ ) = min{
u n (t ) + 2.5anτ (1 − u n (t ) / Vn )(0.025 + u n (t ) / Vn )1/ 2 ,
[
(1a)
]
bnτ + (bn τ 2 − bn 2[x n −1 (t ) − s n −1 − x n (t )] − u n (t )τ − u n −1 (t ) 2 / bˆ )
2
165
}
(1b)
Human Centred Design for Intelligent Transport Systems
where
u n (t ) speed of vehicle n at time t ,
an
n wishes to
maximum acceleration which the driver of vehicle
undertake,
τ
apparent reaction time, the same constant for all vehicles,
Vn
speed at which the driver of vehicle
bn
most severe braking that the driver of vehicle
n wishes to travel.
n wishes to undertake
( bn <0),
x n (t ) location of the front of vehicle n at time t ,
sn
effective size of vehicle
n , that is, the physical length plus a margin into
which the following vehicle is not willing to intrude, even when at rest,
b̂
value of bn −1 estimated by the driver of vehicle
n who cannot know
this value from direct observation.
The position of vehicle
n at time t + τ is thus calculated to be:
x n (t + τ ) = x n (t ) + 0.5[u n (t ) + u n (t + τ )]τ
(2)
The model has been calibrated with the following values: Vn sampled from a
normal population N (20.0, 3.22) m/sec, s n sampled from a normal population N
(6.5, 0.32) m, a n sampled from a normal distribution N (1.7, 0.32) m/sec2, bn
equated to -2.0 a n , b̂ minimum of -3.0 and ( bn -3.0)/2 m/sec2,
3.2
τ
2/3 seconds.
Incorporation of driver behaviour under ISA into Gipps’
model
The analysis of driver behaviour within this study involves the driver behaviour
parameters related to speed (i.e. speed, acceleration and deceleration) that
could be introduced into Gipps’ model. Simulator data for the drives within the
60mph speed limit roads (where vehicle movement is unimpeded) is analysed,
for which the threshold limit for ISA operation was set to be at 62mph.
3.2.1 Speed
First, the impact of the systems on driver speed was analysed. Mean and
maximum speeds were calculated within appropriately defined periods within
which drivers would have already accelerated/decelerated to their desirable
speeds when entering from (or exiting to) different speed limit areas and within
which vehicle movement was unimpeded, as the road sections in which drivers
had to overtake due to lane closure (accident) or slow moving vehicles ahead
166
Modelling of drivers’ behaviour for ITS design
were not included in the analysis. Driver speed behaviour is illustrated in
Table 2.
Table 2. Driver speed under the different ISA systems (mph (m/sec))
Mean
Variance
Maximum
Variance
Base
62.66 (28.01)
5.242 (2.342)
67.13 (30.01)
5.012 (2.242)
Informative
62.71 (28.04)
4.692 (2.102)
68.73 (30.73)
6.872 (3.072)
Warning
59.73 (26.70)
6.092 (2.722)
63.97 (28.60)
6.452 (2.892)
Intervening
59.17 (26.45)
1.652 (0.742)
60.38 (26.99)
1.002 (0.452)
Intelligent speed adaptation systems do affect driver speed both in terms of
mean and maximum values but also in terms of deviation. Driver maximum
speeds are higher than the posted speed limit, and average speeds exceed the
posted speed limit at base condition and under the use of the informative
system. Driving with the use of the informative system results in somewhat
higher maximum speeds (in relation to the base drive); however the differences
are not statistically significant (90% conf. level). Warning ISA results in
considerably lower maximum speeds (statistically significant in relation to the
informative system) and the intervening ISA produces the lowest maximum
values (statistically significant in relation to all systems). In terms of mean
speeds, the informative system produces similar ones to the base condition,
whereas the warning and intervening systems produce quite lower speeds, with
both being below the posted speed limit. Statistical analysis indicated that mean
speeds follow a normal distribution for the base, informative and warning drives,
but not for the intervening ones, which mainly involved similar speeds for the
majority of the drivers with some outliers; this was expected as the speed under
this system is more controlled.
The speed parameter Vn in Gipps’ model involves maximum vehicle speed,
which represents the speed that the driver would adopt when driving under freeflow conditions, which we shall refer to as driver desirable speed. Model
dynamics are such that if there is enough empty space downstream, vehicles
will eventually reach the desirable speed (assuming that the vehicle starts from
being stationary) and will continue cruising at that speed if unimpeded. This
concept of the desirable speed in Gipps’ model does not correspond to the
driver maximum speed in the simulator experiment, as drivers only reach that
speed instantly, but to driver mean speed which describes speed when vehicle
movement is unimpeded. Hence, for the implementation of ISA with Gipps’ carfollowing model the Vn parameter would take the values of mean speed and
their corresponding variance, as presented in Table 2, for each of the simulated
ISA systems.
3.2.2 Acceleration
Vehicle acceleration was also analysed in order to be used as an input in Gipps’
model. Driver maximum accelerations were recorded and averaged over the
number of drivers; the results of this analysis are presented in Table 3.
167
Human Centred Design for Intelligent Transport Systems
Table 3. Acceleration under the different ISA systems (mph/sec (m/sec2))
Maximum
Variance
Base
7.08 (3.16)
4.682 (2.092)
Informative
6.63 (2.96)
4.112 (1.842)
Warning
7.01 (3.13)
3.442 (1.542)
Intervening
6.49 (2.90)
3.252 (1.452)
Drivers employ rather similar acceleration rates with different systems as the
differences are rather small (not statistically significant at 90% conf. level). Still,
the trends indicate that drivers using the intervening ISA demonstrate lower
acceleration rates, and considerably lower deviations, while use of the
informative ISA produces the highest acceleration values. The acceleration
rates between the systems, follow the same pattern as the maximum speeds
(i.e. reduction with system intrusion), indicating that drivers who employ high
speeds also employ high accelerations, thus driving somewhat more
aggressively. Furthermore, statistical analysis indicated that accelerations follow
a normal distribution only for the base condition and informative ISA drive. This
can be attributed to two reasons (a) small sample size and (b) more controlled
behaviour with the increase of system intrusion.
The acceleration parameter used in Gipps’ model is a n and denotes the
maximum acceleration that the driver wishes to undertake. Model dynamics can
be described as follows [13]: a stationary vehicle that starts to move has an
acceleration of 0.3953 a n and continues moving towards Vn with increasing
acceleration up to the point where speed is
0.95
Vn . It then continues
3
increasing its speed towards Vn with lower acceleration. The evolution of
vehicle speed as recorded from the simulator experiment data was somewhat
different. Vehicles demonstrated higher acceleration rates, when their speed
was lower than
0.95
Vn . Hence, the maximum recorded acceleration does not
3
correspond to the model parameter. For this reason, a modified acceleration
was calculated which is the maximum acceleration that was recorded from the
time-step after which vehicle speed exceeded
0.95
Vn (where Vn the driver
3
desirable speed as calculated in Section 4.2.1). The resulting accelerations are
presented in Table 4.
Table 4. Modified acceleration under the different ISA systems
(mph/sec (m/sec2))
Maximum
Variance
Base
4.62 (2.06)
2.382 (1.062)
Informative
4.98 (2.23)
2.392 (1.072)
168
Warning
5.00 (2.24)
2.142 (0.962)
Intervening
4.41 (1.97)
1.212 (0.542)
Modelling of drivers’ behaviour for ITS design
Modified maximum accelerations are substantially lower than the maximum
ones (Table 3), as drivers employ high accelerations at lower speeds. These
differences were found to be statistically significant for all systems except for
the informative one. In addition, accelerations at base condition and under the
informative system follow a normal distribution.
The incorporation of the ISA acceleration in Gipps’ model involves testing which
one of the two estimated maximum acceleration values (a) the maximum
acceleration (Table 3) or (b) the modified maximum acceleration (Table 4) at
base condition produces better model results and then using the corresponding
values for the simulation of the different ISA systems.
3.2.3 Deceleration
To estimate maximum driver decelerations driver behaviour was investigated in
all 60mph road sections, including those containing slow moving vehicles
ahead, road closure and stimulus appearance at the windscreen to which the
drivers were asked to react by reducing their speed to 10mph.
Table 5. Deceleration under the different ISA systems (mph/sec (m/sec2))
Maximum
Variance
Base
-16.18 (-7.24)
2.932 (1.312)
Informative
-15.44 (-6.90)
2.712 (1.212)
Warning
-16.40 (-7.33)
2.832 (1.262)
Intervening
-15.65 (-6.99)
2.632 (1.172)
Driver decelerations appear to be rather similar between the different ISA
systems, and do not follow similar patters as driver speed or acceleration.
Driving under the informative system involves somewhat lower and driving
under the warning system somewhat higher deceleration values. The calculated
deviations are also similar.
In Gipps’ model there are two parameters involving deceleration: one is the
maximum deceleration that the driver wishes to undertake ( bn = -2.0 a n ) and
the second is the maximum deceleration of the vehicle in-front as perceived by
each driver ( b̂ minimum of -3.0 and ( bn -3.0)/2). The observed decelerations
(Table 5) are greater than those calculated by the formulae proposed by Gipps.
Deceleration in Gipps’ model is applied when the movement of the simulated
vehicle is impeded by that of a vehicle in front. It has been found that in the
case of sudden braking by the preceding vehicle the required deceleration to
avoid a crash can be higher than the vehicle’s desired braking or the perceived
deceleration for the preceding vehicle [13]. Hence, the values presented in
Table 5, indicate more extreme deceleration than the ones represented by
parameters bn and b̂ , and subsequently the deceleration parameters used in
the model shall follow model rules and should be provided by the calibration
formulae that Gipps has defined.
As in this study driver behaviour within a 60mph road section is investigated,
the operation of the intervening ISA involving the application of smooth
deceleration values when entering a road section of a lower speed limit with
higher speed, does not apply. It should be noted however, that in the case of
169
Human Centred Design for Intelligent Transport Systems
different speed limit areas driver deceleration should take a fixed value (that
applied by the system) rather than being calculated from the model formulae.
4 Discussion
Intelligent speed adaptation systems affect several aspects of driver behaviour
and their impact varies in relation to the different system HMI functionalities.
The impact depends on the degree of intrusion of the system on the driving
task. In particular, driving speeds, both in terms of maximum and average
values, decrease with ISA systems, and this reduction increases with the
degree of system intrusion. The same pattern is observed with the values of
driver maximum acceleration; however the differences between systems are
considerable lower. Driver maximum deceleration is also slightly affected by the
use of the systems, but no clear pattern emerges from the observations.
The elicited differences between the systems should be introduced in traffic
models, through appropriate parameters. Initially this seems to be a simple task,
but in reality attention should be paid into the (a) the actual meaning of the
model parameters; for example maximum speed in the model corresponds to
average speed as calculated from the observations, (b) other parameter
assumptions; such as parameter distributions, and (c) the calibration of the
model with the new proposed values, as model dynamics may be valid with
somewhat different values than the observed ones.
Future work on this topic involves the investigation of the use of ISA on other
parameters including driver reaction time and vehicle effective size, and ways to
incorporate their impact into Gipps’ model, as well as ISA impact in 30mph and
20mph roads. Following that, the use of ISA will be incorporated in a traffic
model to investigate ISA impact on network conditions.
5 Acknowledgement
The authors are grateful to the European Commission who provided support for
this research and the Transport Research Laboratory for the collaboration
during the driver simulator experiment.
6 References
[1]
[2]
[3]
Spyropoulou, I., Golias, J., Karlaftis, M., Penttinen, M., and Vaa, T.: ‘Risk
factors and intelligent transport system answers – possible opportunities
and shortcomings’. Proceedings of the 11th World Conference of
Transportation Research, Berkeley, June 2007.
Barcelo, J., and Ferrer, J.L.: ‘Microscopic simulation of vehicle guidance
systems with AIMSUN2’. Proceedings of the XIIIth Euro Conference,
Glasgow, 1994.
Leonard, D.R., Though, J.B., Bagueley, P.C.: ‘CONTRAM: a traffic
assignment model for predicting flows and queues during peak periods’,
TRRL Laboratory Report, 1978, R 841, Crowthorne: Transport and Road
Research Laboratory.
170
Modelling of drivers’ behaviour for ITS design
[4]
[5]
Cameron, G.D.B., and Duncan G.I.D.: ‘PARAMICS, parallel microscopic
simulation of road traffic’, Journal of supercomputing, 1996, 10, (1), pp.
25–53.
PTV Vision, VISUM Online – the intelligent traffic platform.
http://www.english.ptv.de/download/traffic/software/VisumOnline_Einleger_2005_e.pdf,
[6]
[7]
[8]
[9]
[10]
[11]
[12]
[13]
Accessed December 2007.
Hidas, P.: ‘Modelling lane changing and merging in microscopic traffic
simulation’, Transportation Research Part C, 2002, 10, (5-6), pp. 351–371.
Hidas, P.: ‘A car following model for urban traffic simulation’, Traffic
Engineering Control, 1998, 39, (5), pp. 300–309.
Liu, R., van Vliet, D., and Watling, D.P.: ‘DRACULA: Dynamic Route
Assignment Combining User Learning and microsimulAtion’. Proceedings
of PTRC European Transport Forum, Coventry, September 1995, pp. 143–
152.
Liu, R., and Tate, J.: ‘Network effects of intelligent speed adaptation
systems’, Transportation, 2004, 31, (3), pp. 297–325.
Gipps, P.G.: ‘A behavioural car-following model for computer simulation’,
Transportation Research Part B, 1981, 15, (2), pp. 105–111.
Gipps, P.G., and Wilson, B.G.: ‘MULTISIM: A computer Package for
Simulating Multi-Lane Traffic Flows’. Proceedings of the 4th Biennial
Conference of Simulation Society of Australia, Queensland, August 1980.
Crosta, D.A.: ‘Parallel SIGSIM: version 3.0 User guide. Working Paper’,
Centre for Transport Studies, 1999, University of London.
Spyropoulou, I.: ‘Gipps car following model - an in-depth analysis’,
Transportmetrica, 2007, 3, (3), pp. 231–245.
171
SESSION 4: TOOLS AND
METHODOLOGIES FOR ITS
DESIGN AND DRIVERS
AWARENESS
173
Tools and methodologies for ITS design and drivers awareness
A SERVICE-ORIENTED SYSTEM ARCHITECTURE
FOR THE HUMAN CENTERED DESIGN OF
INTELLIGENT TRANSPORTATION SYSTEMS
Jan Gačnik, Oliver Häger, Marco Hannibal (DLR, Germany)
jan.gacnik@dlr.de
ABSTRACT: The requirements of modern advanced driver assistance systems
(ADAS) imply not only technical challenges but also create the need of a human
centered development approach, which is necessary to create a unique and
useful assistance. A human centered approach itself has high requirements for
the technical systems in use, flexible functional development is necessary,
features and concepts change quickly. On the other hand, the prototype
systems need to be reliable to maintain safe experiments, especially when
these experiments are being deducted in a car in a real-world traffic situation.
The approach of the Institute of Transportation Systems, German Aerospace
Center (DLR-FS), provides a way to maintain a safe but flexible development
process using service-oriented architecture (SOA) techniques.
1 Introduction and motivation
The introduction and development of modern intelligent transportation systems
(ITS) require a human-centered development approach. This is due to the high
degree of automation which enters modern cars, making it necessary to find
correct ways to assist the driver in his driving-related and non-driving-related
tasks. To achieve this, technical prototypes not only need to be designed with
the human-centered approach, but also need to be evaluated at all stages in the
development process, as early as possible. These continuous evaluations are
necessary to be able to integrate the new knowledge into the next prototyping
iteration. This involves simulation and real-world traffic studies.
The institute operates the research vehicle ViewCar® to analyse drivers‘
behaviour in real traffic situations, determine causes for driver errors, and
conclude in concepts for new assistance systems. Virtual reality laboratories
can be used to evaluate driver assistance functions at an early concept stage
and to observe driver behaviour in controlled scenarios, so that also rare or
dangerous situations may be analysed. Prototype systems can later be
evaluated in a motion based driving simulator, creating a higher degree of
realism. Finally these prototype systems can be evaluated in real traffic using a
specially equipped research car (FASCar). This is equipped with state-ofresearch sensors and actuators making it possible to perform autonomous and
semi-autonomous driving (assisted driving).
To be able to handle the different requirements that these different types of
infrastructure and the corresponding development focus impose, a special
system architecture needs to be defined to support the human-centered
development process in the best way possible.
175
Safety requirements
Human Centred Design for Intelligent Transport Systems
Fig.1. DLR-FS research facilities and requirements
2 Concept
2.1
Requirements for a system architecture
The special requirements that the human-centered development process
imposes are founded in the heterogeneity of the requirements during the
development process itself. Figure 1 visualizes these requirements, in the case
of the DLR facilities.
The advantage of a simulation is the ability to reproduce certain, specific traffic
situations and to deduce experiments without risking accidents. This is
especially important for the assessment of new safety-critical systems.
Designing a system for the simulator needs to be flexible in the first place;
changes in the HMI design or the functionality itself need to be easily possible.
To rapidly evaluate visual HMI displays, tools like RaScal and CoEDiT have
been developed in the DLR-FS [1].
In an advanced stage of the development the same system will be transferred
from the simulation into a real car. For those experiments in real traffic, other
aspects become very important for the design of a system architecture. The
major requirement in real world traffic experiments is safety, which normally can
be neglected in a simulated environment. Up to today only few special
requirements for safety of automotive systems exist, compared to the
aerospace and railway sector. But the upcoming ISO 26262 (Road vehicles –
Functional safety) [2, 3] will define automotive safety-integrity-levels (ASIL),
similar to those in the railway sector. These definitions will have restrictions for
the development process and system architecture and therefore also delivering
a frame for the functions, sensor and actuator configurations.
176
Tools and methodologies for ITS design and drivers awareness
A way needs to be found to support the developer in the best way possible to
support developing cross-platform (and cross-requirement) applications, aiming
to achieve a developer centered process, supporting him in creating drivercentered applications.
2.2
System architecture concept
The technical system architecture itself consists of two aspects, software and
hardware architecture. In this paper we will focus on the software aspects,
because this part is present in both the simulation and the real car. Hardware
architecture is also addressed by this approach but the resulting hardware
architecture is merely a result of an optimization process. It is a combination of
fulfilling a certain functional quality (functional requirements) at a given
acceptable failure level (safety requirements), minimizing economical costs.
e
Se
rv
ic
e
Se
rv
ic
e
Se
rv
ic
e
Se
rv
ic
e
Se
rv
ic
e
ce
i
rv
Se
ce
e
ic
rv
Se
i
rv
Se
e
ic
rv
Se
i
rv
Se
e
ic
rv
Se
ce
Se
rv
ic
Software modules in a vehicle, as well as in a corresponding simulation, are
modeled as part of a distributed system, offering services to other modules (and
to the driver in the end). The system uses a hierarchic model (figure 2)
aggregating services in domains and providing tree-like structure for domain
data.
Fig.2. Domain model containing services
In a domain, services fulfill certain tasks. These tasks can be a recognise-act
cycle or a control loop, depending on the domain. Meta-information describes a
service itself and also the data being exchanged between the modules. All
services, their requirements and orchestration, as well as the domain data
models are described formally using an XML scheme. This description can be
edited using a GUI based application. The formal description allows not only
creating different views on the system or parts of it, but also establishes a very
high degree of portability as most of the platform dependent code can be easily
generated from it.
From a computer science viewpoint, the software architecture concept has
many similarities with service-oriented architecture (SOA) concepts [4, 5], today
often used in modern web applications. One basic idea is the separation of
functional code and formal interface description, including functional and non177
Human Centred Design for Intelligent Transport Systems
functional requirements. Similar as proposed in the rich component model
(RCM) concept [6], a service offers certain functionality and information with a
certain quality and failure rate (promises). On the other hand, another service
might need this functionality and information for its own operation (demands).
The meta-information regarding the service’s reliability can be used online or
offline for analysis, optimization and verification.
To be able to handle safety-critical services as well as more flexible
applications, different runtime environments (RTE) are used. Furthermore,
dependent on the target platform and the requirements static or dynamic data
models are provided. The system is capable to manage different types of RTE
and data models simultaneously, introducing scalable dynamics of the RTE as a
design option.
Due to requirements from safety standards, safety-critical services are
implemented in a very static environment. The code generation process
encapsulates the access to static data on electronic control units (ECUs).
Administrative services, including inter-service operation and control, are very
lightweight on such systems.
Unlike safety-critical services, higher-level services need a more dynamic
environment. This applies for infotainment services, car-to-mobile
communications, or even assistance systems that support a traffic participant
beyond single transportation systems - mobility assistance. Like in conventional
SOA approaches, services can be physically distributed among the system.
The concept of loose coupling [7] is realized by an encapsulated IO interface
that provides the system communication. Once a service is created, other
services can reuse this service and its output data for further purposes.
Furthermore, to provide the specified system behaviour within the scope of a
defined system status, a runtime administration concept is part of the
architecture. Therefore services are being registered during system initialisation
(or if possible during runtime) at the domain master controller (DMC). This
controller guarantees the correct execution of participating services. The DMC
is able to revive services which are in an undefined state. The observation of
given system resources allows system reorganisation by redistribution of
services, for example, in the case of resource breakdown. Of course, due to
hardware restrictions (e.g., actuator control) in some domains this functionality
can only be provided for the dynamic parts of the system.
The whole development process for driver assistance services is an iterative
workflow consisting of the following steps:
•
Specifications and requirements for new ADAS functions are gathered
from the conclusions of ViewCar® studies, but also from literature and
ideas or experiences from earlier developments.
•
Even without programming skills, a service can be described and
specified. The required data can be modeled by a scientist using a GUI
application, which handles XML schemes.
178
Tools and methodologies for ITS design and drivers awareness
•
The service’s functionality is implemented by an engineer following the
given specification in a programming language suiting the specific
domain (C/C++, SCADE, MATLAB/SIMULINK etc.).
•
An automated process provides the availability of the service in all
facilities. This is done via code-generation from the formal XML
scheme, creating the needed view on data, depending on the target
RTE.
•
Furthermore, the new service is tested and parametrised in its
environment.
•
The system is being evaluated, possibly resulting in a further refinement
or new ideas for ADAS.
The process is visualized in Figure 3.
Fig.3. Iterative development process for new ADAS
Because of the very versatile description of services and modeling of data, not
only driver assistance systems can be developed with this approach, but also
simulation modules. These services include modules like vehicle dynamics
simulation and traffic simulation as well as 3D graphics and sound.
3 Evaluation
3.1
Concept implementation
A first prototype of the architecture concept was implemented. This includes
most parts of the different RTEs. As services, all modules of the simulation were
integrated using this concept, as well as some driver assistance functions, like
an adaptive cruise control (ACC).
3.2
Sample ADAS prototype: ACC
The functional behaviour of an ACC system and its requirements is well-known
and standardised [8, 9]. A custom ACC version was developed within the scope
of this project, serving as the base functionality for new adaptive ACC
prototypes which will be developed at DLR-FS. The new versions include,
among others, an ACC that is adapting to the current traffic density [10]. These
prototypes will change the behaviour of distance and speed controllers. The
base ACC is available in the simulators and the real car.
179
Human Centred Design for Intelligent Transport Systems
3.3
ACC realisation
ro
l
on
t
C
ee
d
Sp
e
ic
ce
vi
Se
r
Control Model Task
Tr
af
f
S e ic D
rv en
ic s
e ity
l
Recognise Act Cycle
tro
DataCore
Domain Vehicle
Control
rv
Se
e
on
al
r
Se
vi
c
C
n
di
ce
rv
i
n
ta
itu
Se
is
ng
DataCore
Domain HMI
D
Lo
ce
Following the idea of service definitions, the classic ACC itself consists of
multiple reuseable services which offer longitudinal control over the vehicle.
These services were realised according to the architecture concept and
development process (figure 3). In the case of ACC the functional requirements
are mostly standardised. From these requirements several services could be
identified (speed control, distance control and the corresponding HMI) and
formally described using the XML scheme. Afterwards the service
implementation was done. Figure 4 depicts the identified services and their
domains.
Fig.4. Domain model, ACC specific
The first ACC prototype was developed for the simulator, evaluating the basic
functionality. The next step was the functionality test in a real car on a test track,
thus making migration and parameter fitting very easy. The final step was the
use of the same function in a more robust, safe and static RTE within the car,
for test drives deducted in real traffic. The development of the basic longitudinal
control was a good case study for the evaluation of the interoperability between
different research facilities. During the development the architecture concept
served its purpose very well and confirmed its suitability. Because most of
DLR’s infrastructure itself is part of research projects, changes in the facilities
occur very often. This includes the change of hardware parts (ECUs, sensors,
actuators) as well as software parts. Due to the formal specification of services,
all of these changes could be made with minimal impact on the ACC
development. Once an application programming interface (API) in the simulator
changed or a new ECU was available in the FASCar, a change within a single
template file of the code generator resolved all dependencies.
In further ADAS development the already developed services will be reused to
support new services. For example, as mentioned earlier, a traffic density
awareness service will directly influence the behaviour of the distance and
speed control of the vehicle.
180
Tools and methodologies for ITS design and drivers awareness
4 Conclusions and Outlook
In this paper we presented a concept for a system architecture with SOA
aspects. It provides an integrated approach for the human-centered
development process of ADAS. By redefining the ACC in the form of services,
the suitability could be demonstrated in a proof of concept implementation with
the major focus on the interoperability of different laboratories of the DLR.
The next steps will include further refinement of the architecture, when more
ADAS will be implemented. Safety-critical functions, like a collision-avoidance
system, might require hardware redundancy concepts, meeting safety
standards. Other scenarios like the integration of car-to-mobile and car-to-car
communications will be used to extend and demonstrate the more dynamic
parts of the architecture, including management of reconfigurable services,
aiming at supporting individualised assistance systems and vehicle-independent
mobility assistance. Furthermore the integration of specialised industry tools will
be evaluated, especially for certifiable development of safety-critical
applications.
5 References
[1]
Häger, O., Guraj, V., et al. : ’CoEDIT und RaSCal : Werkzeuge für die
Versuchsvorbereitungen von MMS-Simulatorstudien’, Proceedings of
BWMMS 07, 2006
[2] DIN EN 61508 : ‘Functional safety of electrical/electronic/programmable
electronic safety-related systems’, 2005
[3] ISO WD 26262 : ‘Road vehicles – Functional safety’, 2005
[4] Humm, B., Voß, M., et al. : ‘Regeln für serviceorientierte Architekturen
hoher Qualität’, Informatik-Spektrum, Vol. 29(6), pp. 395-411, Heidelberg,
December 2006
[5] Richter, J.-P., Haller, H., et al. : ‘Serviceorientierte Architektur’, InformatikSpektrum, Vol. 28 (5), pp. 413-416, Heidelberg, October 2005
[6] Damm, W., Votintseva, A., et al.: ‘Boosting Re-Use of Embedded
Automotive Applications Through Rich Components’, FIT – Foundations of
Interface Technologies, 2005
[7] Josuttis, N. M: ‘SOA in Practice: The Art of Distributed System Design’,
O’Reilly, 2007
[8] ISO 15622 : ‘Transport information and control systems – Adaptive Cruise
Control systems – Performance requirements and test procedures’, 2002
[9] ISO 15623 : ‘Transport information and control systems – Forward vehicle
collision warning systems – Performance requirements and test
procedures’, 2002
[10] Knake-Langhorst, S., Schießl, C. : ’Local Traffic Condition – Improvement
of a Vehicle Based Measurement Approach’, ITS 2007 Proceedings, 6th
European Congress and Exhibition on Intelligent Transport Systems and
Services, 2007
181
Tools and methodologies for ITS design and drivers awareness
ANALYSIS AND MODELLING METHODS FOR THE
DESIGN AND EVALUATION OF AN ECO-DRIVING
SYSTEM
Jérôme Barbé, Guy Boy (EURISCO International, France)
jerome.barbe@eurisco.org
ABSTRACT: The modification of driving behaviour can generate fuel
consumption reduction. From this point of view, our solution was to design a
new automotive onboard system to modify the driving activity. This paper
presents the framework adopted to analyse the domain of driver activity in order
to define the preliminary specifications of the system design and evaluation.
1 Introduction
The driving task has become more difficult because the driver has to manage
increasing traffic complexity on the road and various onboard systems
integrated in modern vehicles. The designer of new systems has to take into
account the variability and the multiplicity of parameters concerned in order to
design systems where the cooperation with the driver and the distribution
between these various tasks will respect the context of use. It is important to
consider the interaction between the human and the machine in a dynamic
environment such as the automotive domain in order to reduce human and
system errors during the design process. Cacciabue and Hollnagel [1] make a
review of main existing driver models. For these authors the most useful models
for the analysis and the design are the models correlated with the paradigms of
the vehicle and the environment (Driver-Vehicle-Environment: DVE). However,
the authors specify that to use and select the adequate model, a preliminary
analysis needs to be carried out to define the various objectives in terms of
system design and evaluation. Consequently, designers need an analytical
framework that enables understanding interaction problems with respect to
domain specific parameters. Generally, models are too precise or are too
targeted on the description of specific interactions or particular driving
behaviours. These models do not provide a general analysis framework and an
overall picture of the problem but give a specific answer to a clearly defined
objective.
During the Global EneRgy management and driver Interface for a Citizen
Optimal driving behaviour (GERICO) project a preliminary analysis was done
using a specific framework to understand correctly the domain activity in order
to specify the design and the evaluation of a new eco-driving system. The goal
of this project was to contribute to the reduction of fuel consumption and CO2
production by optimising on-board energy and designing an appropriate
interface that enables the driver to adopt the best driving behaviour, smooth
speed and appropriate gear management. A framework adapted from the
AUTOS pyramid [2] was used to perform the preliminary analysis. This
183
Human Centred Design for Intelligent Transport Systems
framework called “DRIVENTS” includes some existing models to explain some
behaviours or specific interactions in the automotive domain. Using this
framework, the designer can see the interactions between the agents of the
domain studied and especially the links between these agents in order to
understand and define the relevance of each one in the design and the
evaluation of the human-machine interaction.
In this paper we present the DRIVENTS framework adopted to analyse the
automotive domain. According to this analysis, we defined the objectives in
terms of human-machine cooperation in order to select the adequate
requirements for system design and evaluation. Analysing the domain activity
(e.g. automotive domain) allows to take into account dynamic and contextual
parameters and associated concepts. This approach considers that the analysis
of the domain activity was used as a base to define the design model of the
system in order to improve driving performance, comfort and safety of the
driver.
2 Automotive domain analysis : DRIVENT framework
The DRIVENT framework (Fig.1) enables four main agents to connect in the
automotive domain: the DRIver, the Vehicle and onboard systems, the
ENvironment and the driving Task/activity. Each agent was defined by some
characteristics. Drivers, and more generally humans, are characterized by the
variability of their behaviour (inter-individual and intra-individual variance). In
this sense, several factors can influence driver behaviour and are variable for
each driver. Personal factors play a fundamental role in the driving activity: age,
sex, driving experience, social attitudes, cognitive capacities or handicap.
Situational factors depend on the driver’s psychological state at a given time.
These factors are fatigue, motivation, stress, vigilance and emotions for
example. The road environment is variable and complex and can be
characterised by three principal components to define the road context: traffic
conditions, climatic conditions and road infrastructure. The tasks and the activity
represent the primary driving task and the secondary tasks associated (e.g.
radio, phone…). Lastly, the vehicle can be characterised according to its type
(4x4, compact…), its onboard systems and more generally by all its
characteristics (weight, engine…). The analysis of the various relations between
these four agents during the design process is multidirectional in the sense that
each entity influences the other or involves specific relations that the designer
has to consider in the automotive domain before the system specification.
Fig.1. DRIVENT Framework
184
Tools and methodologies for ITS design and drivers awareness
¾ Driver ↔ Tasks: There are three interconnected levels proposed by
Michon to define the driving task [3]: operational, tactical and strategical. The
operational level is related to vehicle control and concerns very short-term
actions (this level can be defined as skill-based according to the Rasmussen
terminology). The tactical level is related to vehicle guidance and concerns
short-term actions (rule-based). The strategical level is related to itineraryfollowing and concerns long-term actions (knowledge-based). Moreover, two
other tasks were carried out by the driver during the driving activity. The tasks of
driving indicators management (speedometer, engine, oil…) and the secondary
tasks that correspond to manage different instruments like the radio, the
onboard computer, the navigation system, but also instruments added in the
vehicle cockpit such as the mobile phone for example. These systems are not
essential for the driving task and provide information which should not disturb it.
These different tasks involve the mobilisation of drivers’ cognitive resources.
Humans have limited capacities or limited attention resources. Driver attention
corresponds to cognitive processes: as perception, access in memory,
reasoning, control, decision-making, and actions [4]. The driver must share
his/her attention and mobilize resources between these various tasks. The
process of time-sharing represents the driver’s capacity to maintain this
attention between the various processes necessary to make the decisions and
provide the adequate actions. Applied to the previous driving task analysis the
driver must share these cognitive processes between five tasks, resulting in the
limitation of available resources. These concepts are important for the designer
because the addition of new onboard systems can increase the mobilisation of
the drivers’ cognitive resources.
¾ Tasks ↔ Vehicle: The driver interacts with the vehicle via the controls
and with the various onboard systems that form an integral part of the vehicle.
The driver is responsible for the management of these various systems. In
some situations, this can result in a possible overload according to the contexts
of use. The workload generated by the accumulation of tasks can have an
impact on driver performance. Thus, it is important that the designer of new
systems makes sure that operator workload is maintained on an acceptable
level. Independent systems integrated in the vehicle can generate interferences
with the driving tasks and can be an important source of distraction. Indeed, in
double task situations, interference or competition for the resources between
concurrent tasks need to be taken into account by the designer. According to
Wickens [5], two tasks that share the same requirements of resources will be
carried out in time-sharing with less effectiveness than two tasks that do not
overlap. Thus, the designer has to select appropriate inputs/outputs and
manage the role and priority of the different tasks to avoid possible coverings.
¾ Driver ↔ Environment: Situation awareness is defined by Endsley [6]
as “the perception of the elements in the environment within a volume of time
and space, the comprehension of their meaning and the projection of their
status in the near future”. Endsley defines three dependent hierarchical levels
preceding decision-making and the action in dynamic situations: perception,
comprehension and projection. Endsley’s model is influenced by the external
factors of the task such as systems which can influence situation awareness
from their interface, interaction level, complexity, and by individual factors
specific to each person, such as experience or training. The designer has to
185
Human Centred Design for Intelligent Transport Systems
develop the system to reduce or to improve the impact on driver situation
awareness. The driver is responsible for his/her mobility and his/her
environmental impact on pollution, traffic or on the other road users. According
to this point of view, the social interaction in the automotive domain is also a
factor where the good citizenship of the driver plays an important role. For this
reason, the interactions between the drivers are an integral part of the driving
activity and can explain some driving behaviour. Onboard systems can have an
impact on the improvement or the deterioration of the relation between the
various agents of the environment. For example, the use of mobile phones or
navigation systems can modify the driver performance (deceleration,
approximate trajectory) and consequently impact on the other road users.
¾ Driver ↔ Vehicle: The link between the driver and the vehicle can be
seen in two ways: the first is the direct interaction with vehicle controls and
onboard systems and the second is the relation between the driver and his/her
vehicle, for example paying more attention in town to avoid small collisions or
on regular maintenance to reduce the risk of breakdowns. These types of
behaviour can be considered safer while the opposite can contribute to
increasing unsafe driving behaviour. The vehicle gives sensory feedback
indications to the driver on acceleration, deceleration or the aspect of the road
for example. This information is necessary to provide the driver with driving
sensation. The concept of driving sensation can also be considered according
to the vehicle type. The driving behaviour will be different if the driver has a
sports or a city vehicle. Thus, it is necessary to take into account the
driver/vehicle couple and the relation that exists between them to study and to
understand some driving behaviour.
¾ Environment ↔ Tasks: The driver analyses environmental situations
and takes adequate actions according to a given context. The environment can
generate specific interactions and modify the task and the activity of the driver.
The driver will react consequently while adapting to the external conditions such
as climatic or traffic condition for example. This type of interaction is more a
direct perception-action cycle and can correspond to the driver’s automated
responses. We can take the example of an emergency action in a very short
time like emergency braking or obstacle avoidance.
¾ Environment ↔ Vehicle: The relation between the vehicle and the
environment can be seen by physical interactions (tires for example) but also by
some onboard systems which communicate with the environment (electronic
pricing or GPS). In the near future, it will be important for the designer to
consider the possibilities of communication between the environment and the
vehicle, generally called “Car to Infrastructure Communication” and
communication between vehicles called “Car to Car communication”.
3 System design analysis : DRIVENTS framework
At the first stage of the analysis, four agents and their main relations were
defined by the DRIVENT framework for the automotive domain. In the second
stage, the DRIVENTS framework purposed to show the relations and the
concepts linked to the integration of a new system in the vehicle in order to
186
Tools and methodologies for ITS design and drivers awareness
define the requirements for the system design and evaluation (Fig.2). The
designer has to take into account the different agents to have a clear
understanding of different parameters and variables involved in these relations.
Fig.2. DRIVENTS Framework
The DRIVENTS framework (Table.1) resumes the different concepts coming
from the DRIVENT analysis and the concepts used to design a new system to
cooperate with the driver, the driving activity, the vehicle and the environment.
Table 1. DRIVENTS matrix (Top-down lecture)
Driver
Vehicle
Environment
Task/activity
System
Dri
Driving
sensation
Situation
awareness
Cognitive
resources
mobilisation
Interface
Ergonomic
Relational
aspects
V
Communication
Infrastructure
Vehicle & on
board systems
management
Integration
Environment
Social
aspects
Mechanical
interactions
En
Situated actions
Interaction &
communication
Task/activity
Task &
activity
analysis
Role &
priority
Reaction
T
Prescribed
tasks
Procedure
Interaction
Coordination
Contextualisation
Training
S
Driver
Vehicle
System
The DRIVENTS matrix shows the different parameters involved to design a new
system such as integration, coordination, contextualisation, training, prescribed
tasks, interface and interaction that the designer has to consider to improve the
human-machine integration according to the concepts defined in the DRIVENT
preliminary analysis, e.g. with respect to the agents of the automotive domain.
In this chapter, we give an overview of different parameters that we took in
consideration for the design of an eco-driving system in relation with the
different concepts defined in the DRIVENTS framework.
¾ Eco driving system principle: The goal of the eco-driving system was
to change the driving task to reduce fuel consumption. The GERICO system
187
Human Centred Design for Intelligent Transport Systems
interacts with the other agents defined in the DRIVENT framework. The relation
existing between them was modelled to describe agent interactions (Fig.3). The
system developed was based on a partial analysis of the environment and did
not determine and anticipate all road situations. Indeed, the automation of some
system functions was not an optimal solution to aid the driver in his/her driving
task. The system only provides information and warning to assist the driver that
keeps the control of the vehicle. The principal problem in the interaction was the
choice between intrusiveness/severity of the system to keep an optimal driving
performance and a good acceptability for the comfort and safety of the driver.
Fig.3. Agents interactions
The system provides prescribed tasks via an HMI to reduce fuel consumption
with respect to contextual and dynamic constraints of the automotive domain. A
global optimisation algorithm takes into account the environmental data via the
navigation system (road type, topography, traffic), and internal data from onboard computers (fuel injection, engine speed). This algorithm processes these
data to provide the driver with an economic driving pattern (prescribed tasks) for
a given trip (Fig.4).
Fig.4. GERICO principle
188
Tools and methodologies for ITS design and drivers awareness
¾ System ↔ Tasks: In the analysis of the eco-driving principles, the
speed and gear management were two main factors to be considered for the
reduction of fuel consumption. Consequently, the Eco-driving system provides
prescribed tasks, e.g. optimal speed and gear advices, to the driver. Some
information concerning eco-driving activity will not be provided by the system
(for example optimal acceleration or deceleration). The system will aid the driver
for some driving tasks but other tasks will be entirely performed by the driver to
adapt his/her driving activity in some situations. In this sense, it is humanmachine cooperation with an allocation of the cognitive functions between the
system and the driver. However, the driver remains the person in charge of the
human-machine couple to adopt the best eco-driving activity with respect to the
road context. We can consider that the first impact of the system was on the
operational level because the system provides information influencing directly
the control of the vehicle (speed and gear advice) and consequently influences
the other levels. These changes were considered during the design process
and observed during evaluations. The social impact of this system was also
important to consider. This system acts on the “driving style” and tries to change
driver habits. From this perspective, the driver’s motivation to use the system is
crucial. Consequently, system usefulness needs to be demonstrated in terms of
fuel consumption reduction, as well as usability and acceptability.
¾ System ↔ Driver: The driver interacts with the system via a specific
interface. The ergonomic questions concerning the interface are important to
facilitate the human-machine interaction. For Norman [7, p61] "User-centred
design emphasizes that the purpose of the system is to serve the user ... The
needs of the users should dominate the design of the interface, and the needs
of the interface should dominate the design of the rest of the system". Thus, the
interface of the system was developed using many guidelines such as ISO
(International Standards Organisation), AAM (Alliance of Automobile
Manufacturers) guidelines and so on, a user-centred and participatory design
approach during the design process to improve perceived complexity and
affordance of the system with respect to driver needs in order to avoid human
errors and dysfunction of the “Human-Machine coupling”.
Fig.5. GERICO instrument cluster view
A multimodal interface [8] was developed based on the analysis of the possible
interferences with the driving task. Four types of information were provided by
189
Human Centred Design for Intelligent Transport Systems
the system with two modalities: Navigation (visual and auditory), Assistance
(auditory), Advice (visual and auditory) and Warning (visual and auditory). All
visual information was proposed via an instrument cluster design (Fig.5). The
auditory information was implemented in addition to the visual modality to
provide information to the driver. This modality enables better information
management to avoid excessive workload in visual perception that can
generate disturbances at the level of the main driving task.
¾ System ↔ Environment: Dey [9, p4] gives the following definition of
the context: “context is any information that can be used to characterize the
situation of an entity. An entity is a person, place, or object that is considered
relevant to the interaction between a user and an application including the user
and applications themselves”. This author defines that “a system is contextaware if it uses the context to provide relevant information and/or services to the
user, where relevancy depends on the user’s task” [9, p5]. Thus the designer
has to take into account as well as possible the situational parameters of the
road context during the driving activity such as climatic conditions, traffic
conditions, infrastructure and specific events that can occur on the road.
Consequently, optimal advices, warning and assistances information are
contextualised with respect to the road context. The system developed
diagnoses some driving activities and determines the mode of interaction and
the most adequate type of information in the situational context. Even if we send
the right information, the way it is delivered is as important as the information
itself. Information sent to the driver has to be useful and well synchronised with
the car’s position and driving context (essentially road topography and
infrastructure). For example, the system provides limited information in town
due to driver workload and cognitive resources mobilisation. Moreover,
anticipation of situations can improve driving performance to reduce pollution as
well as safety. Thus, the system provides assistance information to anticipate
some elements of the road infrastructure such as right of way, dangerous
curves and zones limited to 30 km/h for example.
The Interaction Control System (ICS) was developed to provide the driver with
visual information (optimal speed and gear advices) and vocal messages with
respect to road context in order to reduce cognitive interferences, risk and
human errors. A multi-agent model was used [10], based on cognitive function
analysis, to define the structure of the ICS (Fig. 6). The ICS includes several
specific agents that take into account internal and external parameters involved
in the driving task. Agents are involved in Driving Activity Assessment (DAAs),
in Driving Activity Analysis (DAAn), in Environment Assessment (EA) and in
Multimodal Management (MM). DAAs agents control the driving activity (speed,
gear and pedals), DAAn agents compare the global driving activity with the
optimal driving pattern, EA agents analyse the road context and MM agents
manage visual and auditory information provided to the driver.
190
Tools and methodologies for ITS design and drivers awareness
Fig.6. Interaction Control System architecture
¾ System ↔ Vehicle: The system integration was controlled in function
of the technological limits of the vehicle and onboard systems integrated. The
system developed needs resources from other systems (vehicle and GPS data).
It was necessary to set up communication and reliable coordination in order to
manage the roles and the priorities with the other systems. Information was
provided to the driver via the LCD (liquid crystal display) instrument cluster
(fig.5) and vocal messages via the vehicle radio.
4 Conclusion
This paper presented the “DRIVENTS” framework used to analyse the driving
activity context during the design process of a new eco-driving system. This
framework enables the specification of the design of the system with respect to
the agents involved in the automotive domain. This framework can associate a
systemic approach to the problem and an anthropocentric approach. It can be
used for a simple system or a complex system in order to improve comfort,
performance and safety. Moreover this framework was used to identify and
define variables for the system evaluation [11]. The framework showed that
different driver’s factors can be influenced by the addition of new systems in the
vehicle such as workload, situation awareness, time-sharing, cognitive
interference and mobilisation. The questions concerning ergonomic,
communication, interaction or interface aspects need to be resolved with
respect to the driver’s needs in order to reduce the influence of the new system
on the agents in the automotive domain defined in the DRIVENT framework.
Finally, during the design process, the designer has to verify as well as possible
that all points were considered and evaluated to improve the system design in
order to obtain the best human-machine cooperation with respect to the context
of use.
191
Human Centred Design for Intelligent Transport Systems
5 References
[1]
Cacciabue, P. C., and Hollnagel, E.: ‘Modelling driving performance: A
review of criteria, variables and parameters’. Proc. Int. Conf. Modelling
driver behaviour in automotive environments, Ispra, Italy, Mai 2005, pp. 2527
[2] Boy, G.: ‘Human modelling program: initial approach and literature review’.
Eurisco Technical Report, 2005
[3] Michon, J.A.: ‘A critical view of driver behaviour models: What do we know,
what should we do?’, in Evans, L. & Schwing R.C., (Ed.): ‘Human behavior
and traffic safety’. (New York: Plenum Press, 1985), pp. 485-524
[4] Amalberti, R.: ‘Le manque de ressources au centre des débats’, in Le
travail Humain (Ed.): ‘La conduite des systèmes à risques’ (PUF, Paris,
1996), pp. 66-75
[5] Wickens, C.D.: ‘Attention, Time-Sharing and Workload’, in Wickens, C.D.
(Ed.): ‘Engineering psychology and human performance’ (Harper Collins
publishers Inc, 1992, 2nd edn.), pp. 364-411
[6] Endsley, M. R.: ‘Toward a theory of situation awareness in dynamic
systems’, Human Factors, 1995, 37, (1), pp. 32-64
[7] Norman, D. A.: ‘Cognitive engineering’, in Norman, D. A., & Draper, S. W.
(Ed.): ‘User centered system design: New perspectives on humancomputer interaction’ (Hillsdale, NJ: Lawrence Erlbaum Associates, 1986),
pp. 32-65
[8] Barbé, J., and Boy, G.A.: ‘On-board system design to optimise energy
management’. Proc. Int. Conf. European Annual Conference on Human
Decision-Making and Manual Control, Valenciennes, France, September
2006
[9] Dey, A. K.: ‘Understanding and Using Context’, Personal and Ubiquitous
Computing, 2001, 5, (1), pp. 4-7
[10] Barbé, J., and Boy, G.A.: ‘An Agent Interaction Model for System Design’.
Proc. Int. Conf. Interest and use of Joint-Cognitive Models for Human
Centred Design, Vienna, September 2007
[11] Barbé, J., Boy, G.A., and Sans, M.: ‘GERICO: A Human Centred EcoDriving System’. Proc. Int. Conf. 10th IFAC/IFIP/IFORS/IEA Symposium on
Analysis, Design, and Evaluation of Human-Machine Systems, Seoul,
Korea, September 2007
192
Tools and methodologies for ITS design and drivers awareness
AN ANALYSIS OF THE REQUIREMENTS OF
DRIVER ASSISTANCE SYSTEMS – WHEN AND
WHY DOES THE DRIVER LIKE TO HAVE
ASSISTANCE AND HOW CAN THIS ASSISTANCE
BE DESIGNED?
Julia Werneke (DLR, Germany), Astrid Kassner (DLR, Germany), Mark Vollrath
(Braunschweig University of Technology, Germany)
Julia.Werneke@dlr.de
ABSTRACT: The development of Advanced Driver Assistance Systems (ADAS)
strongly depends on engineering prospects [1]. In addition to that, users’ needs
are a basic prerequisite in the development process of ADAS to increase users’
acceptance [2, 3]. The two main goals when developing ADAS are increasing
driving safety and comfort [4]. But what are additional motives for the usage of
ADAS? Within a focus group 12 experienced drivers (four women, eight men)
discussed why and in which critical driving situations the driver would like to be
assisted. Seven motives were identified: driving safety, comfort, speediness,
flexibility, driving pleasure, health, and saving of costs. Driving safety, comfort,
and driving pleasure are the core motives for “good” driving. The study gives
indications that besides driving safety and comfort there are other motives
which have to be considered or can be potential starting points for the
development of novel ADAS.
1 Background
Traditionally, the development of Advanced Driver Assistance Systems (ADAS)
strongly depends on technical possibilities [1]. The driver often does not have
an active role in the development of such systems. Consequently, there are
many ADAS technologies available on the market which are not readily adapted
to the needs of the driver. According to Kollmann [5], not every system that is
technically possible ensures success in the users’ acceptance and in sale.
Therefore, not every driving task which can be supported technically by ADAS
has to be implemented on a system. In addition to the technical limits, the
needs of the user are also a basic prerequisite in the development process. A
product like an ADAS is accepted and demanded from the driver, when both the
user’s need exists and the product is adapted to that need [2]. According to
INVENT, involving potential users in earlier stages of development ensures
close connection to the requirement and acceptance of systems [6]. Thus, when
developing ADAS the driver should be involved early in the development
process, e.g. in assessing driver needs for ADAS.
Increasing driving safety and comfort are the main goals when developing
ADAS [4]. The task of ADAS is to support or relieve the driver in “normal”
driving, e.g. with navigation and speed control, but also to support or counteract
mistakes of the driver when they lead to an accident, e.g. ESP and ACC [4].
193
Human Centred Design for Intelligent Transport Systems
The question is, whether there are other criteria in addition to driving safety and
comfort which should be followed when developing ADAS. To answer this
question, the general aim of this study is to determine additional factors that
influence the driver in the driving situation and are therefore important for the
driver during driving. According to Fuller’s “Task-Capability Interface model of
the driving process (TCI)”, driving is defined as “the management of task
difficulty” [7, p. 48]. Certain factors influence whether the driver has a driving
situation under control: Driving is the interaction between (1) the demand of the
driving task to achieve a safe outcome and (2) the capability applied during the
task (0).
The “task demands” (D) are the objective parts of driving and contain features
of the environment, the driving behavior of other road users, and the
characteristics of the vehicle, like its speed and road position. The task
demands represent the dynamic element in the model. The “capability” (C),
which is the subjective part, contains the driver’s limit of competence his or her
skills. These competences are, e.g. control skills, hazard detection, recognition,
anticipatory, and defensive driving skills. The capability refers to “what the driver
[…] is able to do at any given moment” [7, p. 49]. It indicates the momentary
ability of the driver to deliver his or her level of competences. Consequently, the
competence sets the limit on capability. But the capability is also determined by
a range of factors, e.g. fatigue, emotion, alcohol and other kinds of drugs,
stress, and distraction.
Fig.1. The Task-Capability Interface model [7]
The model (0) shows that the control of the driving situation develops based on
the task demands and the momentary driver’s capability. When the capability
exceeds the task demands (C>D), the driver has control about the driving
situation and s/he will manage the driving task. The driver will consider the
driving task as easy and probably boring. When the task demands exceed the
capability (D>C), the driver will consider the driving task as difficult. The
194
Tools and methodologies for ITS design and drivers awareness
capability is beyond the scope of the driver’s competence to manage the task
demands of the driving situation. The result is losing control which may result in
a crash. But also when the capability equals the task demands (C=D), the driver
can perceive the driving task as stressful or annoying [7].
Thus, one driving situation, in which ADAS can provide assistance, are
situations where the task demands exceed or are clearly below the driver’s
ability. Developing assistance systems to support drivers in these situations
might lead to a decrease of drivers’ strain and an increase in comfort resulting
in an increase in drivers’ safety.
The aim of this study was to find out what is important for the driver during
driving and why s/he would like to be assisted. The study was also designed to
determine driving situations – with regard to the purpose of the journey, and
physical and environmental conditions – that the driver evaluates as difficult and
where s/he would like to be assisted. This study focuses on driving situations in
which the driver is unable to manage the situation with his or her momentary
driving skills (C<D). The need for driver assistance should be indicated by
difficult driving situations (1st step). The study should also indicate how
assistance systems could be designed in those driving situations to support the
driver (2nd step).
To get this information, a focus group was chosen. This method gives a first and
wide review about the opinions, values, attitudes, and conflicts of the
participants, in a relatively fast and economic way [8]. A focus group is one of
the methods that can be used to investigate users’ needs, especially for
assistance systems [3]. It provides an insight into up to now unknown or less
structured areas under investigation [8], like the development of novel ADAS. A
focus group enables people “to ponder, reflect, and listen to experiences and
opinions of others. This interaction helps participants compare their own
personal reality to that of others” [9, p. 17]. Thus, the requirements of the
participants according to ADAS can be pointed out.
2 Method
Twelve experienced drivers, four women and eight men, aged between 25 and
47 years (mean= 36.3, sd= 7.5) discussed the requirements of ADAS within a
focus group. They focused on driving situations in which they would like to be
assisted and why they would like to have this driving assistance. Furthermore,
the discussion sought to identify how the assistance could be designed. Novice
drivers and drivers over 50 years were excluded to increase the homogeneity of
the subject group.
The discussion was conducted at the German Aerospace Center (DLR) in
Braunschweig. It was recorded by a camera and in writing. The timeframe was
three hours. The discussion was conducted on the basis of a discussion guide
that outlines the concept of the discussion in addition to a “warming-up” and a
“final” phase. The concept and the central issues of the focus group are shown
in 0. The concept was visualized through the metaplan board.
195
Human Centred Design for Intelligent Transport Systems
Concept of the Focus Group
Phase 1
Existing Means of Transportation
Phase 2
Motives
Central Issues
How can the assistance
be designed?
Why does the driver
like to have assistance?
Phase 3
Purpose of the Journey
Physical Conditions
When does the driver
like to have assistance?
Environmental Conditions
Fig.2. Concept and central issues of the focus group
The approach of the focus group is based on three phases. These phases
address the three central issues of the study (0). The first phase – “means of
transportation” – addresses the central issue how the driver would like to be
assisted. Existing means of transportation ought to provide information about,
how the assistance can be designed. It contains, e.g. someone driving for you
(e.g. taxi), others driving with you (e.g. bus, train), or others helping during the
driving (e.g. co-driver, driving instructor). Compared to driving alone, driving
with companions includes persons who assist or make the driving pleasant. It
represents a potential approach which can be compared with driver assistance.
In the second phase – “motives” – the goal is to determine the advantages and
benefits of these collected means of transportation that the driver would like to
have in his or her own car. The second phase represents the core module of
the focus group. It ought to answer the central question why the driver would
like to have assistance. Consequently, the motives provide information about,
what is important for the driver. The motives show which needs for driver
assistance potentially exist and where the development of ADAS should start.
The third phase – “purpose of the journey, physical and environmental
conditions” – ought to address the central issue when the driver would like to be
assisted. In this phase difficult driving situations are mentioned and the needs
for driver assistance will be specified. The third phase is subdivided into three
categories: (a) purpose of the journey, (b) physical conditions, and (c)
environmental conditions. The purpose of the journey represents the definite
reason for driving. According to Benda [10], the purpose of the journey
distinguishes between driving to work, running errands, driving in leisure time,
and going on vacation. The physical and environmental conditions are
according to Fuller [7] the factors of capability and task demands that influence
the driving situation, e.g. tiredness and stress of the driver or weather
conditions, other road users, vehicles, or road signs.
3 Results
The focus group was transliterated by camera and protocol. The spoken words
with participant number were transliterated. This transcript provided the basis of
196
Tools and methodologies for ITS design and drivers awareness
the interpretation of the data according to the three central issues (0). The
interpretation occurred reductively, which means that the quantity of the data
was reduced with the goal to gain some compact statements [8]. The
interpretation occurred according to the concept of the focus group (0). For the
analysis, the statements are considered in terms of the group and not in terms
of a single participant [8].
The phase “motives” is the basis for the interpretation of the statements. The
number of statements regarding the motives is put into parentheses. The
mentioned “means of transportation” point out whereby the assistance can be
compared. They will also be named in parentheses. The discussion showed
that regarding the “purpose of the journey”, the participants referred to the
motives. This general aspect reflects approaches why the driver would like to be
assisted at different purposes of driving. According to the specific aspects of
driving – “physical and environmental conditions” –, the connection to the
motives could not be shown in the discussion. In fact, the participants described
system properties of assistance systems in special driving situations. In this
paper only the difficult driving situations due to environmental conditions are
described.
According to the participants, seven motives out of 112 statements are
important for “good” driving (0a). The most important motives are driving safety
(36 statements) and comfort (31 statements). In addition to them, the
participants mentioned speediness (13 statements), flexibility (6 statements),
driving pleasure (6 statements), health (5 statements), and saving of costs (4
statements). The terms of the motives were deduced from the statements of the
participants. In this paper we focus on the five motives mentioned most
frequently.
(a)
(b)
Requirements of ADAS in Driving Situations
due to Environmental Conditions
Motives of the Requirements of ADAS
45
Driving Safety
36
35
31
Number of Approaches
of ADAS
Number of Statements
40
30
25
20
13
15
10
11
6
6
5
5
4
0
Safety
Comfort
Speediness
Flexibility
Driving
Pleasure
Health
Saving of
Costs
Others
16
14
12
10
8
6
4
2
0
Comfort
13
3
1
Environment
1
Road User
2
2
Road
Fig.3. (a) Number of statements in regard to the motives of the requirements of
ADAS. (b) Frequency of ADAS addressing the motives driving safety and
comfort, divided into environment, other road users, and road
The motive “driving safety” is considered most important. It is divided into safety
with regard to the vehicle and to the driver. The vehicle itself constitutes safety
due to the crush zone (e.g. bus, plane). Safety also means to have safe road
contact, e.g. on black ice. Driving safety with regard to the driver consists of his
or her skills and competences. The participants award the pilot, the bus driver,
and the driving instructor the highest competences. Especially, the taxi driver
197
Human Centred Design for Intelligent Transport Systems
gets the highest skills in driving situations where the driver is unable to drive
due to alcohol or has to drive in an unknown town. Getting feedback about the
knowledge of the traffic rules, such as from the driving instructor, is also
accepted by the larger part of the participants. Thus, driving safety does not
exist when the driver is unable to cope with the driving task. On the one hand
this can be a consequence of environmental conditions (e.g. rain, fog), on the
other hand due to physical and mental impairment of the driver’s skills (e.g. lack
of knowledge about the route). For the development of driver assistance
systems, it means that the task of ADAS should be to increase driving safety,
but not just by reacting to environmental conditions. Increasing driving safety by
ADAS also means the support of the driver and the improvement of the vehicle
(e.g. crush zone).
“Driving comfort” focuses on the convenience during driving (e.g. in bus, train,
plane, metro, taxi). It would be most desirable that the driver is relieved of the
driving task, so the journey time can be used for other things, e.g. the
passenger can relax or prepare work. Driving comfort also means to reach the
destination with the lowest amount of stress as possible (e.g. in train, rickshaw).
When developing ADAS, the systems should contribute to the well-being of the
driver, so that driving becomes stress-free, relaxing, and exonerative. These
two motives are so important in driving that they are postulated for every
purpose of journey – driving to work, running errands, driving in leisure time, or
going on vacation.
“Speediness” entails the needs to reach the destination fast, on time, and
reliably (e.g. bicycle, metro and over ground in town; car in country side).
Especially, this motive is desired on the navigation level when driving with a
deadline, e.g. driving to work, to the doctor, or to the theatre. If speediness
cannot be provided, driver assistance at the navigation level is useful to make
the driving fast and reliably.
“Flexibility” is very relevant in routing. Dropping the search of parking places
represents flexibility for the driver (e.g. bicycle, motor cycle). This motive is
especially desired in driving to work and going on vacation. On the one hand
the driver enjoys mobility, on the other hand s/he wants to increase his or her
individuality. Assistance systems should not restrict the driver in his or her
manner of driving and in routing.
Additionally, according to the participants, driving is evaluated as “good” when
driving is fun (e.g. bicycle, motor cycle, rickshaw, riding a horse). “Driving
pleasure” is a basic motive, like driving safety and comfort, which is mentioned
independently of the purpose of the journey. Thus, ADAS should not restrict the
driving pleasure or reduce the driver’s activity. Driving should be active and
suspenseful for the driver, so that a “positive” driving feeling is maintained.
With regard to the physical and mental conditions of the driver, assistance
systems are desired when the driver is mainly stressed, tired, overstrained or
unchallenged through the driving task, distracted, surprised, aggressive,
frustrated, appalled, and alcoholised. These conditions strain the driver. ADAS
should avoid these physical conditions and should open up possibilities to
counteract them.
198
Tools and methodologies for ITS design and drivers awareness
According to the environmental conditions, the participants did not refer to the
motives. They described system properties of driver assistance systems in
special driving situations. There was no association to the motives mentioned,
but the mentioned driver assistance systems reflect the motives driving safety
and comfort (0b).
0b shows the number of desired approaches of assistance systems in driving
situations the driver evaluates as difficult. The situations are divided into three
conditions: environment, other road users, and road. In the focus group for
almost every mentioned driving situation, the participants gave an approach of
assistance systems that are desired in that situation.
Out of 24 mentioned driving situations, 22 desired approaches of assistance
systems described the aim to improve the driving safety and comfort. 0b shows
that more assistance systems (AS) are mentioned that increase driving safety
(18 approaches of AS), compared to comfort (4 approaches of AS). Especially,
driving safety (13 approaches of AS) is desired in situations characterized
through the environment, e.g. weather conditions, time of day, or temperature.
The same is shown in the category “road user”, e.g. traffic flow and traffic
density (3 approaches of AS of driving safety and 1 approach of AS of comfort).
In the category “road” (e.g. road network, road surface) in each motive two
assistance systems are described that aim at providing driving safety or
comfort.
According to the results of the category ”environment”, the driver would like to
be assisted in weather conditions like rain, snow, fog, and sun (7 of 13
approaches of AS for increasing driving safety) and darkness (6 of 13
approaches of AS). Especially, the reduced detection of the road and of the
traffic in front, due to fog or darkness, are evaluated as difficult. The bad
visibility and thus the reduced orientation on the road overstrain the driver. The
direct glares due to the sun or its reflection in other vehicles are also mentioned
as difficult. It constrains the driver in his or her manner of driving. Another
critical driving situation is seen in the slip hazard on black ice or on snowcovered and watery roads. In all these situations, the driver desires a driving
assistance system which clarifies the driver’s view and provides a safe road
contact of the vehicle (“an assistance system that shows on the basis of
arrowhead the course of the road” VP 311; an assistance system that informs
the driver about the current weather conditions” VP 040). In the focus group the
sudden appearance of wild game at night is also evaluated as critical and in
need of assistance. In those driving situations, the driver desires an assistance
system with anticipatory perception of wild game and warning of the driver. In
regard to the driving at night, one participant mentioned the desire of an
assistance system that drives autonomous, so the driver will be relieved (1
approach of AS for increasing comfort).
According to the results of the category “road user”, the participants evaluate
driving with low traffic flow or traffic density as comfortable (“you feel free, …
you do not feel so hemmed in” VP 231). But in low traffic density in combination
with bad visibility conditions (e.g. due to rain, fog, darkness), the driver would
like to be assisted by a system that supports the driver’s view and relieves
him/her (1 of 3 approaches of AS for increasing driving safety). Driving safety is
also desired in driving situations with traffic jam or a sudden braking vehicle in
199
Human Centred Design for Intelligent Transport Systems
front (2 of 3 approaches of AS). The driver desires assistance systems that
warn him/her about those unexpected situations. But in traffic jam, the driver
would also like to be assisted by stop and go to make the driving more
comfortable (1 approach of AS for increasing driving comfort). Drivers desire an
assistance system that controls the driving task completely (“so you do not have
to put the foot the whole time on the clutch” VP 231).
According to the results of the category “road”, a bad road quality (e.g. cobble
streets, loose chippings streets) is seen as a risky driving situation (2
approaches of AS for increasing driving safety). Automated driving assistance
systems, which interfere appropriately regarding the conditions of the road
surface, are desired so the car is not running off the road. Besides the
automated lateral control, automated assistance is also desired in regard to
braking so that a safe braking is provided by the assistance system. Driving in
an environment with a high number of traffic signs is seen as risky, too. In the
view of the participants, it leads to excessive demands of the driver (2
approaches of AS for increasing comfort). Especially, when the routes are
unknown, this type of driver overload is mentioned. A driving assistance is
needed that reduces this excessive demand. The assistance system should
perceive the traffic signs and inform the driver about the current driving
situations.
Table.1 gives an overview about the mentioned approaches of ADAS in driving
situations due to environment, other road users, and road.
200
Tools and methodologies for ITS design and drivers awareness
Table1. Overview about the desired approaches of ADAS in critical driving
situations due to environmental conditions
Environmental
Conditions
Environment
rain, snow,
fog, and
darkness,
sun
snow, rain
darkness
Road User
traffic density
traffic flow
and density
other road
users
Road
Road surface
Road network
Driving Situation
Task of ADAS
reduced detection of the road
and of the traffic in front
- Clarifying the driver’s view (S)
- Informing the driver about the current
weather
conditions and the course of the road (S)
- Autonomous driving (darkness) (C)
direct glares, reflection in other
vehicles
slip hazard on black ice or snowcovered and watery roads
wild game
low traffic density in combination
with bad weather conditions (e.g.
rain)
(unexpected) traffic jam
sudden braking vehicle in front
bad road quality (e.g. cobble
streets, loose chippings streets)
high number of traffic signs
- Providing a safe road contact of the
vehicle (S)
- Anticipatory perception of wild game and
warning the driver (S)
- Clarifying the driver’s view (S)
- Warning the driver about the traffic jam (S)
- Automated stop and go driving (C)
- Warning the driver about the braking (S)
- Automated interfering appropriately
regarding the conditions of the road
surface (S)
- Perceiving the traffic signs and informing
the
driver about the current driving situations
(C)
S= approaches of ADAS for increasing driving safety; C= approaches of ADAS
for increasing driving comfort
4 Discussion and Conclusion
The focus group shows that “good“ driving involves different criteria. In addition
to the traditional criteria of driving safety and comfort, there are other criteria like
speediness, flexibility, and driving pleasure that are also important for “good”
driving. When these criteria are fulfilled, the driver will perceive driving as
optimal. However, when some criteria are not fulfilled in different driving
situations, these criteria can be starting points for the development of ADAS.
The focus group confirmed the results of Timpe [4] that the two most important
criteria for the driver and thus for the requirements of ADAS are driving safety
and comfort. Both are desired in every kind of journey. The discussion shows
that especially driving safety contains not only aspects of environmental
conditions; also the driver and the vehicle influence the safety of driving.
Although in the discussion a connection to the motives was not made according
to the “physical and environmental conditions”, the mentioned approaches of
assistance systems reflect the two motives. The approaches show that
especially the desire for driving safety is shown in driving situations
characterized through the environment (e.g. weather conditions or darkness;
201
Human Centred Design for Intelligent Transport Systems
and through other road users, e.g. traffic jam or a sudden braking vehicle in
front). Besides safe driving, comfortable driving is also important for the driver.
The discussion shows that assistance systems should not restrict the driving
pleasure. It is a core motive of “good” driving and is desired in every kind of
journey. The criteria speediness and flexibility are special motives that are
desired in driving to work and going on vacation. For that purpose, driving
should be as fast and as on-time as possible, with a routing that is adapted
individually to the driver.
The focus group is an established method for the investigation of users’
requirements, especially for assistance systems [3]. But a general problem of a
focus group is the incomplete amount of data [8]. Because of the discussion
situation, the individual opinions and attitudes of every participant cannot be
investigated. In this regard, the subjective data motives of the participants can
just be assumed from the investigator. Other problems of the method are
opinion leaders and socially desired answers [8]. Because of the group
situation, participants may not dare to say their opinions. In following studies the
method should be extended, e.g. with interviews or standardized
questionnaires. The weakness of the sample is the small number of participants
(N= 12). Thus, the results cannot be generalized and should be handled with
care. But a focus group is a perfect method for the generation of ideas and for
quickly gauging user opinions about a topic, like the requirements of ADAS [3,
8]. The review about the drivers’ attitudes is useful for further research. Thus, in
following studies a higher number of focus groups should be conducted, so that
more representative data is available.
The study showed how experienced drivers can contribute to the development
of ADAS. The focus group showed the focus of requirements of driver
assistance systems and the different approaches when ADAS are desired.
ADAS should not only be designed according to external influencing factors of
the driving task like environment. The systems should also take into account
both the physical conditions of the driver and the vehicle. It is important that
assistance systems contribute to driving safety and comfort. But driving
pleasure is another important criterion which is desired from the driver in every
driving situation. Speediness and flexibility are very situational motives for driver
assistance. However, these could also be possible starting points when
developing novel driver assistance systems. The focus group focused on the
long-term aspects for the requirements of ADAS in driving. An additional aspect,
which could be put into focus, is the short-term aspect – the driving maneuvers.
By this, it could be examined which driving maneuvers are seen as difficult.
Such a group discussion could reveal in which driving situations driving
assistance is desired from the driver and why it is desired.
5 References
[1]
Beier, G., Boemak, N., and Renner, G.: Sinn und Sinnlichkeit –
psychologische Beiträge zur Fahrzeuggestaltung und -bewertung, in
Jürgensohn, T., and Timpe, K.-P. (Ed.): Kraftfahrzeugführung (Springer,
2001), pp. 263-284
202
Tools and methodologies for ITS design and drivers awareness
[2]
Christ, S., and Baur, A.: Die Rolle des Fahrers bei der Entwicklung neuer
Infotainment- und Fahrerassistenzsysteme im Automobil, in Verein
Deutscher Ingenieure (Ed.): Der Fahrer im 21. Jahrhundert,
Anforderungen, Anwendungen, Aspekte für Mensch-Maschine-Systeme.
VDI-Berichte Nr. 1768 (VDI, 2003), pp. 257-268
[3] Wandke, H., and Wetzenstein, E.: Assistenz für Benutzer: Den Umgang
mit Technik erleichtern (Humboldt Universität, 2003)
[4] Timpe, K.-P.: Fahrzeugführung: Anmerkungen zum Thema, in Jürgensohn,
T., and Timpe, K.-P. (Ed.): Kraftfahrzeugführung (Springer, 2001), pp. 9-27
[5] Kollmann, T.: Das Konstrukt der Akzeptanz im Marketing: Neue Aspekte
dargestellt
am
der
Akzeptanzforschung
Beispiel
innovativer
TelekommunikationsMultimediasysteme,
und
Wirtschaftswissenschaftliches Studium, 1999, 28, (3), pp. 125-130
[6] Schierge, F.: Welche Fahrerassistenz wünschen sich die Fahrer?
Methoden und Ergebnisse zur Nutzerakzeptanz von FAS aus der
Forschungsinitiative INVENT, in VDI Wissensforum (Ed.): Fahrer im 21.
Jahrhundert, Der Mensch als Fahrer und seine Interaktion mit dem
Fahrzeug. VDI-Berichte Nr. 1919 (VDI, 2005), pp. 207-219
[7] Fuller, R.: The Task-Capability Interface model of the driving process,
Recherche Transports Sécurité, 2000, 66, pp. 47-57
[8] Lamnek, S.: Gruppendiskussion: Theorie und Praxis (Beltz, 2005, 2nd
edn.)
[9] Krueger, R. A., and Casey, M. A.: Focus groups: A practical guide for
applied research (Sage, 2000, 3rd edn.)
der
Gefährlichkeit
von
[10] Benda,
H.
v.:
Die
Skalierung
Straßenverkehrssituationen, 1. Teil: Ein Klassifikationssystem für
Verkehrssituationen aus Fahrersicht (Technische Universität München,
Lehrstuhl für Psychologie, 1977)
203
Tools and methodologies for ITS design and drivers awareness
A HUMAN FACTOR-BASED APPROACH FOR THE
EFFECTIVE USE OF DRIVING SIMULATORS AND
E-LEARNING TOOLS IN DRIVER TRAINING AND
EDUCATION
Jose M Pardillo Mayora (Technical University of Madrid, Spain)
jmpardillo@caminos.upm.es
ABSTRACT: The present state of driving simulator technology makes it possible
to implement driver training applications with a growing level of complexity and
fidelity to real driving conditions. However, driving simulators only become an
effective tool in driver training if they are effectively included as an integral part
of the training curriculum. Such integration requires a methodical approach as
well as a detailed analysis of the training curriculum, the learning goals and
training needs. This paper presents the results of Humanist Task Force G
regarding a definition of a human factor-based approach for the effective use of
technology in training taking into account cognitive aspects, as well the special
needs and requirements of the training process for different target groups:
novice drivers, professional drivers, elderly drivers, and disabled drivers.
Methodological conditions for enhancing the acquisition of different levels of
driving abilities and for using new technologies as a tool of performance
measurement are discussed,
1 General framework for driver training
It is generally accepted that driver training must include a theoretical and a
practical component, which ideally should enable the student to acquire the
required abilities to master the driving task by means of an abstract approach
and through practical training by driving in a real vehicle in real traffic.
Driver training curricula are based on theoretical assumptions about driver
behaviour and the driving task. A common premise is that driver behaviour is
organised in three hierarchical levels –strategic, tactical and operational – as
described by Keskinen [2], building on previous work by Michon [2,3]. Based on
this work, the European project GADGET described the issues relevant for
driver training and in particular training of novice drivers in order to provide a
theoretical framework to define and cover the areas of competences that driver
training and education need to address [4]. This project provided the basis for
developing the Goals for Driver Education (GDE) framework.
The GDE-framework comprises two dimensions. The first dimension includes
the three hierarchical levels mentioned above. A fourth level concerning “goals
for life and skills for living” was added. It refers to personal motives and
tendencies in a broader perspective than simply conducting the driving task.
This level is based on the assumption that lifestyle, social background, gender,
age, income, etc. have an influence on attitudes, driving behaviour and accident
involvement.
205
Human Centred Design for Intelligent Transport Systems
In addition to the four levels of driver behaviour, the GDE-framework operates
with another dimension constituted by three goals for training: knowledge and
skill, risk-increasing factors, and skills for self-assessment. These dimensions
all relate to behaviour at the different levels thus influencing both driving
preconditions as well as the accomplishment and execution of the driving task.
The knowledge and skill level refers to the skills a driver needs for driving under
different circumstances.
Risk-increasing factors deal with aspects of driving or traffic that may increase
the risk, such as perception of the traffic situation, speed adjustment, and risk
acceptance.
The skills for self-assessment refer to the driver’s capability to assess his
performance on the four behavioural levels. It really points to critical selfadjustments of everything from skills in vehicle handling to a reflection of
individual risk attitudes.
Ideally, driver training curriculum should cover all the areas of the framework
and address the appropriate driving behaviour associated with them [5].
However, traditional driver training typically only covers the last two levels of
behaviour (vehicle manoeuvring and mastery of traffic situations). Similarly, at
the goal dimension level only the first level of knowledge and skills is included
and risk increasing factors are covered to a lesser extent in existing training
programs. The third level is rarely encountered during driver training.
2 Target groups
2.1
Novice drivers
Novice drivers need to acquire both theoretical and practical skills as well as
knowledge at all levels of the GDE-framework.
Novice drivers are the group with the highest accident risk. Fatality rates for 18
to 24-year old drivers (number of driver fatalities per million age group) in
industrialised countries are about the double of those of experienced drivers
(aged 25 to 54). During the first years after having passed the driving test, the
accident risk declines sharply. Several studies have found that the accident risk
decreases rapidly during the first years of driving experience. However, it takes
about 7 years of driving experience before the accident risk reaches an
acceptable, low level [6].
In-depth accident analysis of crashes involving young novice drivers has
revealed that it is not so much a lack of basic driving skills that has caused the
accident, but a lack of so-called higher order skills. These skills deal with risk
perception, risk acceptance, self-assessment, the motivation to drive safely, etc.
Young drivers tend to misinterpret traffic situations and show an inefficient
visual search. Furthermore, they have difficulties in adapting their speed and
driving distance to the driving conditions [7]. They also tend to underestimate
the risk of a hazard resulting in an accident and overestimate their ability to deal
with hazards [8].
206
Tools and methodologies for ITS design and drivers awareness
There are indications that simulator training helps to speed up the learning
process, although at present the knowledge on transfer and retention of the
skills acquired during simulator training is insufficient to assess its effects on the
performance of the drivers after the training period [9].
2.2
Professional drivers
Professional drivers typically undergo training in handling difficult driving
manoeuvres and driving special vehicles or goods. Simulators have been
developed for both research and training of professional drivers. Already in
1958 the Iowa State driving simulator linked a vehicle cabin mock-up to a
scaled physical terrain model allowing the driver to control actions in a
rudimentary road layout. Since then there have been many different technical
innovations including videos of real scenes and, more recently, computer
generated environments. At present, several European countries employ truck
driving simulators as part of in-service training.
Brock et al. studied the effects of training with three types of simulators that are
often used in the US to retrain experienced drivers: an open-loop video
simulator, a low-end simulator, and a mid-range simulator [10]. The use of
simulation decreased trainee drop-out rates by 35% for an agency using the
midlevel simulator, decreased student failure rates by 50% in an agency using
the open-loop and the low-end simulators, and decreased the collision rate by
10% in an agency using a combination of open-loop and low-end simulators. In
addition, the use of simulation reduced training time in one agency from 19 days
to 17 days by replacing classroom bus training with simulator training. In
another agency using only the open-loop system, training time was reduced by
5 days when simulation was employed.
Strayer and Drews conducted an experiment with professional drivers who
spent 2 hours in a simulator learning shifting strategies in order to maximise fuel
efficiency [11]. The participants’ fuel consumption performance when driving
their normal route in their own vehicles was monitored for 6 months. Training
increased fuel efficiency by an average of 2.8% over the six-month interval.
Drivers who drove not specifically simulated vehicles in the training sessions
were also found to improve their performance after training.
In the UK, Parkes and Reed conducted a longitudinal cohort study that sought
to provide an analysis of the benefits of synthetic training in the area of fuel
efficiency improvement [12]. 36 drivers received training designed to improve
their driving style in a range of traffic situations in a truck simulator
During each simulator visit, apparent fuel consumption figures were recorded
and compared to real-world fuel consumption records for the same drivers. In
addition and over the same period, fuel consumption data were obtained for a
matched cohort of drivers who did not attend the training. The mean change in
fuel efficiency observed of the drivers in the simulator group showed an
improvement of 15.7 % in their on-road performance.
These results suggest that simulators can be effectively used to improve
specific driving skills in experienced drivers.
207
Human Centred Design for Intelligent Transport Systems
2.3
Elderly and disabled drivers
Similarly, training groups of elderly or disabled drivers would have to implement
different approaches such as focusing on the functional awareness of their
abilities and limitations and adapting their driving behaviour accordingly without
exposing them to the risks of real traffic.
Roenker et al. (2003) designed a study to examine if a speed-of-processing
training can improve at-risk older adults’ driving performance. Training transfer
and retention were assessed in an on-road driving evaluation immediately after
training and 18 months after training. Simulator training resulted in an
improvement in the specific driving manoeuvre skills that were expressively
practiced during training. However, those effects were mostly temporary, and
dissipated at the 18-month follow-up [13].
Disabled drivers are typically trained in order to identify functional weaknesses,
i.e., not for the purposes of training per se but rather to be able to identify the
driving conditions in which they would encounter difficulties. In this context
simulators can be of great use in helping the subjects to gain functional
awareness of their abilities and to adapt their driving behaviour to them without
being exposed to the risks of real traffic [14].
Akinwuntan et al. designed a study to test the effectiveness of simulator training
in heart stroke patients [15]. Eighty-three first-ever subacute stroke patients
entered a 5-week 15-hour training program in which they were randomly
allocated to either an experimental (simulator-based training) group or a control
(driving-related cognitive tasks) group.
Simulator-based driving training
improved driving ability, especially for well educated and less disabled stroke
patients. However, the findings of the study may have been modified as a result
of the large number of dropouts and the possibility of some neurological
recovery unrelated to training.
Simulation can also be a useful tool for assessing drivers with disabilities.
Driving aids could be simulated so that drivers could test them together with the
assessor in a safe artificial environment.
3 Including simulators in the training curriculum
Driving simulator technology makes it possible to implement driver training
applications with a growing level of complexity and fidelity to real driving
conditions. Driving simulators only become an effective tool in drivers’ training if
they are effectively incorporated as an integral part of the training curriculum.
Such integration requires a methodical approach and a detailed analysis of the
training curriculum, the learning goals and training needs [16]
A simulator is an abstraction of reality, and many aspects cannot be reproduced
with sufficient detail or realism. The term fidelity refers to how closely the
simulation imitates reality. From an educational point of view the learning goals
determine the required level of fidelity. Limited fidelity restricts the range of
tasks or task aspects that can be trained in the simulator.
Slick et al. found that transfer of training from simulated environments to the
real world was maximized when training is characterized by a high degree of
208
Tools and methodologies for ITS design and drivers awareness
both physical and psychological fidelity [17]. Nevertheless, some research
results indicate that higher fidelity does not always improve the training results
[18]. Beginning trainees could be overwhelmed by the complexity of the real
system and environment and may, therefore, sometimes be better served by a
simplified, lower fidelity simulation [19]. More experienced trainees, however,
would learn more from a high-fidelity simulation [20].
3.1
Basic vehicle handling skills
Simulators can be used for training in the first steps of vehicle handling. The
advantages are related to safety - trainees would be able to learn these skills
without endangering themselves or other road users - and to environmental
preservation since the use of simulators instead of practice on a motor vehicle
eliminates the pollutant emissions created by the latter.
Trainees should learn to recognise or experience closely the risk-increasing
aspects of the tasks, especially underestimation of speed. By enabling trainees
to evaluate their skills in a realistic way, they will learn to compare their
estimates with the real outcome.
3.2
Cognitive aspects and decision process
Novice drivers lack perceptual skills and anticipation in traffic. On the one hand
they do not use peripheral vision, and on the other hand they underestimate the
time needed for many manoeuvring tasks such as overtaking, merging, lane
changing, reaching an intersection, stopping, and turning. They have problems
estimating the behaviour of other road users as well, i.e. how much time these
drivers need to perform the tasks mentioned above. When an unexpected and
unusual situation does occur, they do not know how to react adequately.
Recent developments in software make it possible that drivers in a simulator
can learn to behave in very realistic way. Automatic Traffic Generation and
Autonomous Driver models reproduce the circumstances in real traffic and
enable users to repeat and therefore train certain tasks in changing
environments, with varying risk, and different road users, with variable
behaviour. These devices facilitates training of anticipatory skills, like risk or
hazard perception, which are highlighted by recent research as very important
for safe driving [21].
The main advantages of simulators compared with real driving are that trainees
can experience scenarios which are too dangerous to create on the road, and
that they can train cognitive skills without fully automated manoeuvring skills.
In order to increase driving skills without increasing excessively the selfconfidence of the trainees, the manoeuvring component should not be
overemphasised. It is preferable that the training process includes
demonstrations and exercises in which novice drivers may fail in order to
develop a realistic self-evaluation of their capabilities. Additionally, trainees
should be given comprehensive feedback on their attitudes, risk perception and
personal tendencies during training. Self-assessment tools like questionnaires
and scales, discussions with other trainees about personal experiences and
209
Human Centred Design for Intelligent Transport Systems
evaluations made by instructors or examiners seem to be appropriate
educational methods.
In a simulator trainees will better experience the harmful influence of factors
such as stress and mood on driving behaviour, and how drivers can cope with
these risk-increasing aspects.
3.3
Training of higher order skills
Training of higher order skills included in the GDE-matrix would mean that the
present driver training should extend the theoretical and practical training
curriculum to areas including confrontation with difficult or risky situations or
with situations where the drivers’ attitudes (motivations, representations and
emotions) are challenged.
However, the sequencing for the acquisition of higher order skills is important.
Higher order skills are primarily integrated through experience and not solely
obtained through education or training. They should be introduced in training
after the trainee completes the skill-and-knowledge based levels.
Groeger casts some doubts on the usefulness of driver simulators for the
acquisition of higher order skills [22]. Simulators make it possible to structure
driver training and to introduce mass repetition of skill drills. Laboratory
experiments have shown that mass repetition of partial tasks helps speeding up
the learning process; but retention is lower. This is in particular the case when
subjects have to conclude from the context what skills have to be applied.
Therefore, in Groeger´s opinion, associative learning during many hours behind
the wheel under varying conditions with an experienced driver next to the
learner driver is a better way of acquiring higher order skills than structured
simulator training in a reduced environment.
Falkmer and Gregersen [23] conducted an experiment to assess the
effectiveness of hazard perception training with simulators in Sweden. Driving
school learners received multimedia training on PC´s and subsequently
simulator training focused on hazard perception and risk acceptance in addition
to their driving lessons. A limited positive effect was observed when using a
simulator with a 120° screen view angle and simple movement feedback. When
a very simple simulator (40° view angle and no feedback) was used, the
simulator training had absolutely no effect on the hazard perception and risk
acceptance by these driving school learners.
These results indicate that the potential of driver simulators for the acquisition of
higher order skills is limited.
4 Conditions for an effective use of simulators in driver
education
Training effectiveness depends on the degree to which trainees are able to
apply the knowledge, skills, and attitudes gained in a training context to the real
context. This means that training is effective when transfer of training is
achieved, and the value of a training device is determined by the degree of
210
Tools and methodologies for ITS design and drivers awareness
learning and transfer that occurs. Barnard et al. [24] defined the following forms
of transfer:
•
Positive transfer: Extent to which trainees have acquired knowledge, skills
and attitudes, which can be applied effectively in work practice.
•
Negative transfer: Extent to which an undesired effect occurs after following
a course.
•
Far transfer: Transfer when the initial learning task and the subsequent
tasks to be learned differ substantially.
•
Near transfer: Transfer when the initial learning task and the subsequent
tasks to be learned differ only slightly or not at all.
•
General transfer: The trainee acquired certain working methods, knowledge
and skills, which can be used in tasks other than the original learning task.
•
Specific transfer: The learning task is so specific that no transfer can be
expected to other tasks, but only to the same task.
The main goal of driver training is to achieve positive transfer: the things
learned in the simulator should be transferred to the real-life driving task. At the
same time it is important to avoid negative transfer. With simulators this is a
possible risk, as the technical driving task is certainly not the same as in a real
car, especially with relatively simple simulators. Perceptive tasks are also
different. For example, the perception of speed and distance is different
compared with real driving.
Although specific transfer is not a problem in itself - the trainee only needs to
transfer the learned driving task to a driving task in the real environment general transfer is desired when it concerns traffic insight. The traffic situations
encountered in the simulator should be generalised to other traffic situations
[25].
Despite their prevalent use, there still exists a lack of evaluation studies to
provide evidence on the effectiveness of driving simulators in driver training.
Humanist Task Force G has synthesised the existing experience on the
effectiveness of simulators as an educational tool [26]. It was found that at
present there is not enough knowledge on transfer and retention of the skills
that are acquired during simulator training to assess its effects on the
performance of the drivers after the training period. Nevertheless, existing
research provides indications of the potential efficiency of simulators and elearning with respect to improving some aspects of driver training [27].
The following recommendations for effective use of simulators in driver training
and education were formulated as a conclusion of Humanist TFG work.
4.1
Novice drivers
Simulator-based training scenarios and instruction program should provide:
•
A valid environment for practicing the necessary skills.
•
Clear goals and contents for training.
211
Human Centred Design for Intelligent Transport Systems
•
Enough feedback to improve behaviour and to learn.
•
A possibility to gain enough experience.
•
A learning period long enough to commit the skills and knowledge learned
to memory, and a learning climate favourable to safety.
It is recommended to extend the curricula in order to cover all the levels of the
GDE-framework. Trainees should learn to know or better experience the riskincreasing aspects of the tasks, especially underestimation of speed. In order to
enable trainees to evaluate their skills in a realistic way, it should be possible for
them to compare their estimates with the real outcome.
In order to increase driving skills without excessively increasing the trainees’
self-confidence, the manoeuvring component should not be overemphasised. It
is preferable that the training process should include demonstrations and
exercises in which novice drivers may fail so that they may develop a realistic
self-evaluation of their capabilities.
Training of novice drivers should include instruction on the proper and safe use
of ITS devices. Discussion and work groups combined with simulator and
multimedia training might be appropriate methods to make novice drivers aware
of specific problems in connection with the use of ITS.
4.2
Professional drivers
Training safe driving strategies can only be successful if driver training covers
motivational and self-evaluative aspects. Trainees should be given
comprehensive feedback on their attitudes, risk perception and personal
tendencies during training.
Physical fidelity is not a sufficient precondition for optimal training effects. For
the training of certain skills (e.g. hazard perception) simulations with lower
fidelity (e.g. video scenes) can be efficient training tools.
4.3
Elderly and disabled drivers
Retraining of disabled older drivers requires the development of specific
scenarios based on their particular needs.
Simulation can also be a useful tool for the assessment of drivers with
disabilities. Driving aids could be simulated so that drivers could test them
together with the assessor in a safe artificial environment.
5 Further research
As a conclusion of the analytical review of driver training and education process
conducted by Humanist Task Force G, research needs were identified in the
following areas [28]:
a) Pedagogical and didactical components
•
Development of effective training curricula (activities, sequence of
activities, scenarios etc) including simulator-based training particularly
212
Tools and methodologies for ITS design and drivers awareness
designed for the needs of specific target groups including novice,
experienced, professional, elderly and impaired drivers.
•
Guidelines for the role of the trainer and his task in the simulator
training process.
•
Criteria for the selection of training media (classroom, slide show,
mechanical devices, computer, e-learning, simulation, real driving, etc.)
for different training activities.
•
Assessment of the effectiveness of using simulators for the acquisition
of higher order skills such as risk perception, and the retention of skills
acquired through simulator training.
b) Technological aspects
•
Technical specifications defining the minimum conditions required for a
simulator in order to be suitable for use at the different levels of driver
training applications.
•
Development of a framework for converting vehicle performance data
into usable training feedback for the driver.
•
Development of a driver performance monitoring tool to assess the
driver's decisions and actions in specific driving situations in real time.
•
Development of an advanced traffic model to produce a realistic traffic
environment around the learner.
•
Techniques to avoid simulator induced sickness.
6 References
[1]
[2]
[3]
[4]
[5]
[6]
Keskinen, E.: Why do young drivers have more accidents? Referate der
Esten Interdiziplinären Fachkonferenz 12-14. Berichte der Bundesanstalt
fur Strassenwesen. Mensch und Sicherheit, Heft M 52, 1996
Michon, J.A.: A critical view of driver behavior models: what do we know,
what should we do? In: Evans, L. and Schwing, R. (Eds.) Human Behavior
and Traffic Safety. Plenum Press, New York, 1985
Michon, J. A.: Explanatory Pitfalls and Rule-Based Driver Models. Accident
Analysis and Prevention, Vol 21,1989, pp. 341-353.
Hattaka, M., et al. : Results of EU- Project GADGET, WP3, 1999.
Gregersen, N.P. et al.: Driving simulator scenarios and requirements –
Deliverable D4.1 GRD1-1999-10024 – UE TRAINER Project System for
driver Training and Assessment using Interactive Evaluation tools and
Reliable methodologies, 2001
Turetschek, C.: Potentials and Risks of New Technologies in the Training
of young drivers. Proceedings of the Conference on European guidelines
for the application of new technologies for driver training and education.
Humanist NoE, Madrid, 2006
213
Human Centred Design for Intelligent Transport Systems
[7]
[8]
[9]
[10]
[11]
[12]
[13]
[14]
[15]
[16]
[17]
[18]
[19]
[20]
[21]
Clarke, D. et al: Voluntary Risk Taking and Skill Deficits in Young Driver
Accidents in the UK. 37:3 Accident Analysis and Prevention, 2005, pp.
523-529.
Deery, H.: Hazard and Risk Perception Among Young Novice Drivers,
Journal of Safety Research,Vol 30, No 4,1999
Vlakveld, W.: What does educational psychology tells us and what are the
implications for simulator training?. Proceedings of the Conference on
European guidelines for the application of new technologies for driver
training and education. Humanist NoE, Madrid, 2006
Brock, J., Jacobs, C., and McCauley, M.: Simulators and Bus Safety:
Guidelines for Acquiring and Using Transit Bus Operator Driving
Simulators. TCRP Report 72, TRB, The National Academies, Washington,
D.C., 2001.
Strayer, D. and Drews, F.: Simulator Training Improves Driver Efficiency:
Transfer from the Simulator to the Real World. Proc., Second International
Driving Symposium on Human Factors in Driver Assessment, Training and
Vehicle Design, Aspen, CO, 2005, pp. 190–193.
Parkes, A. and Reed, N.: Fuel Efficiency training in a full-mission simulator.
Behavioural Research in Road Safety Fifteenth Seminar, Department for
Transport, London, 2005.
Roenker, D. et al.: Speed-of-processing and driving simulator training
result in improved driving performance. Human Factors, 45(2), 218- 233,
2003.
Rizzo, M., et al.: An abstract virtual environment tool to assess decisionmaking in impaired drivers. In J. D. Lee, M. Rizzo & D. V. McGehee (Eds.),
Proceedings of the Second International Driving Symposium on Human
Factors in Driver Assessment, Training, and Vehicle Design. University of
Iowa, Iowa City, 2002.
Akinwuntan ,A. et al.: Effect of simulator training on driving after stroke: A
randomized controlled trial. Neurology , 2005, vol. 65, pp. 843—850.
Pardillo J.M. et al.: Guidelines for the development of a European driver
training and education tool. GUPM-061030T1-AB Humanist deliverable G4
Humanist NoE, Lyon, 2006
Slick, R. et al.: Using Simulators to Train Novice Teen Drivers : Assessing
Psychological Fidelity as a Precursor to Transfer of Training - DSC
Asia/Pacific, 2006 – Tsukuba.
Alessi, S. M.: Fidelity in the design of instructional simulations. Journal of
Computer-Based Instruction, 15(2), 1988, pp. 40-47.
Alessi, S. and Trollip, S.: Multimedia for learning: Methods and
development. Needham Heights, MA, 2001.
Verstegen, D.: Iteration in instructional design: An empirical study on the
specification of training simulators. Doctoral dissertation, Utrecht
University, the Netherlands, 2003.
Espié, S.: Training novice driver: toward cognitive training. Proceedings of
the Conference on European guidelines for the application of new
technologies for driver training and education. Humanist NoE, Madrid,
2006
214
Tools and methodologies for ITS design and drivers awareness
[22] Groeger, J.A.: Understanding Driving-Applying cognitive psychology to a
complex everyday task. Psychology Press Ltd. Hove, 2000
[23] Falkmer, T. and Gregersen, N.P.: The TRAINER Project- The Evaluation
of a New Simulator-Based Driver Training Methodology. In: Dorn, L. (red.),
Driver Behaviour and Training, proceedings of the 1st international
conference on driver behaviour and training, Stratford-upon-Avon,
England, 2003, pp. 317-330.
[24] Barnard, Y., Veldhuis, G., and Rooij, J.: Evaluation in practice: identifying
factors for improving transfer of training in technical domains. Studies in
Educational Evaluation, 27 (3), 2001, pp.269-290.
[25] Shea, J. and Morgan, R.: Contextual interference effects on acquisition,
retention, and transfer of a motor skill. In: Journal of Experimental
Psychology; Human learning and memory, vol. 5, nr. 2, p. 179-187, 1979
[26] Pardillo, J.M. et al.: Report on the effectiveness of simulators as
educational tool. GUPM-071015T1-DA. Humanist deliverable G5 Humanist
NoE, Lyon, 2007
[27] Dolan, D.M., et al.: Simulator training improves driver efficiency: transfer
from simulator to real world. Proceedings of Second International Driving
Symposium on Human Factors in Driver Assessment, Training and Vehicle
design, Park City, Utah, USA, 2003.
[28] Pardillo, J.M. et al.: Functional requirements of driver training and
education tools, identification of research needs and potential applications
of e-learning. GUPM-051220T1-DA (2). Humanist deliverable G3 Humanist
NoE, Lyon, 2005
215
Tools and methodologies for ITS design and drivers awareness
COMPARATIVE EVALUATION OF TRAINING
METHODS IN IMPROVING DRIVERS’
UNDERSTANDING ABOUT THE FUNCTIONALITIES
AND POTENTIAL LIMITATIONS OF ADAS
Villy Portouli (CERTH/HIT, Greece), Evangelos Bekiaris (CERTH/HIT, Greece),
Sofie Boets (IBSR, Belgium), Per Henriksson (VIT, Sweden)
vportou@certh.gr
ABSTRACT: A common study was undertaken to evaluate the effectiveness as
regards knowledge acquisition of three different methodologies for drivers’
training on new in-vehicle technologies, a standard paper manual, a multimedia
software tool and a driving simulator. 93 persons from three countries were
assigned to three groups, the first being trained on four systems by a paper
manual, the second group being trained by a software tool and the third being
trained by a software tool and acquiring practical experience of the system on a
driving simulator. The analysis focused on the correctness of answers given by
the participants on multiple-choice questions. The software group answered
better than the other groups in one of the four systems (Lane Departure
Warning), but no clear effect of the additional training via the simulator was
found. However, people trained with the software and the simulator feel better
trained than the rest. Based on the analysis performed, we can conclude that
the use of software tools and driving simulators for training drivers in the use of
new assistance systems should be further explored.
1 Introduction
1.1
Background
Within Task Force F of the HUMANIST project a new methodology for training
all drivers on the use of new in-vehicle assistance systems has been developed
[1,2]. The work has focused on the use of information technology for this
training, thus a multimedia training tool (MMT) and specific scenarios for a
driving simulator have been realised, the latter aiming to acquaint the driver with
operational specificities of in-vehicle assistance systems.
The assumption is that the use of MMT for training in in-vehicle assistance
systems and the users’ familiarisation with such systems in a driving simulator
will improve the users’ knowledge on in-vehicle assistance systems potential
benefits and limitations [2,3]. It is supposed that drivers usually do not read
detailed manuals and hence, if such a structured curriculum does not exist, they
may misinterpret the in-vehicle assistance systems functionality which may
even result in adverse safety effects, i.e. due to overconfidence in the system or
due to risk homeostasis effects. These potential effects have been analysed in
detail in ADVISORS project Deliverable 3/8.1 [4].
217
Human Centred Design for Intelligent Transport Systems
1.2
Objectives of the study
A common study was performed by three partners of the HUMANIST
Consortium, CERTH, IBSR/BIVV and VTI. The objective of the study was to
evaluate the effectiveness of three training/ learning methods in improving
drivers’ understanding about the functionalities and potential limitations of
Advanced Driver Assistance Systems (ADAS), In-vehicle Information Systems
(IVICS) and Driving Support Systems (DSS).
The three training methods that were tested are training/learning by:
•
Group A: reading a paper manual,
•
Group B: using a specific developed multimedia training tool (MMT),
•
Group C: using the MMT in addition to hand-on experience of each
system.
By comparing the first two groups, one can identify differences between a
simple manual and the MMT. By comparing the second and third group, one
can identify the effect of a simulator compared to the use of an MMT only. The
simulator can not be used as stand-alone training method, as its aim is only to
emphasise the limitations of each system and not to demonstrate all of its
functionalities.
Four systems have been chosen for consideration [5].
Table 1. Systems included in the study
System
Adaptive Cruise Control (ACC)
Lane Departure Warning (LDW)
Navigation system & Route Guidance
Anti-lock braking system (ABS)
Type
ADAS
ADAS
IVICS
DSS
2 Method
2.1
Experimental design
The independent variable was, the training method used, with three levels (A:
training by reading manual, B: training by using the multimedia training tool, C:
training by using the multimedia training tool and providing in addition hand-on
experience). For each training method, all the above mentioned systems were
learned. Providing that, training by reading a manual constitutes the existing
standard training method, this condition was considered as reference (control
condition) for the evaluation of the second condition. Comparison between
second and third condition allowed reviewing any effects caused by the use of
the driving simulator.
218
Tools and methodologies for ITS design and drivers awareness
2.2
Participants
93 drivers from three countries, 33 from Greece, 30 from Sweden and 30 from
Belgium, were randomly assigned in the three experimental conditions,
balancing between gender. Subjects were found through advertisements and
calls. The selection strategy was random selection of novice drivers (having had
a driving license for less than three years), who had at least a sufficient level of
English understanding, trying to have equal men and women drivers in each
group. The sample can be considered as representative of the novice drivers
population as far as gender is concerned. The sample size is rather small, and
this was due to time and resources limitations, thus the results presented in the
following sections should be considered as indicative.
The characteristics per group are given below.
Table 2. Participants’ sample
Group C
Group B
Group A
Mean Age
(years)
27.26
27.10
27.63
Male
15
15
15
Female
16
15
17
Mean
driving
experience
(years)
7.77
7.56
6.51
Mean
Annual
Mileage
(km)
9984
8297
7536
Table 3. Participants’ experience with the systems surveyed
Group C
Group B
Group A
ABS
22
25
19
ACC
2
2
2
LDW
1
0
0
Navigation
system
8
8
8
Driving
simulator
3
1
0
The group participants were matched as regards PC skills, level of English
understanding and educational background. No pre-test for these background
variables was used, due to time and resources constraints. English
understanding was assessed, based on the subjects’ statement. Subjects with
not sufficient English understanding were not able to participate anyway, as
they were not able to read and understand the manual. The majority of the
participants were of university level background, and only few were of basic
education. The participants were assigned to each group, so as to match their
educational background per group, however this fact should be kept in mind in
the following sections.
219
Human Centred Design for Intelligent Transport Systems
Table 4. Group matching
Level of English
understanding
PC use
Expert Average Novice
Group C
Group B
Group A
2.3
5
5
6
25
24
21
1
1
4
High
Sufficient
14
11
15
17
19
17
Educational background
University High
degree school
19
9
15
14
19
11
Basic
3
1
2
Apparatus
All the educative information that was provided to the participants was the
same, irrespectively of the training used. However, depending on the
learning/training method, different kind of apparatus/equipment was necessary.
In condition A (control condition), participants were given a printed manual
providing all relative information in regard to the selected ADAS/IVICS. The
manual was in the form of text, containing a detailed description of systems’
functionality, benefits and limitations in the systems’ use depending on
road/weather/traffic conditions, as well as, some indicative interfaces providing
information about the type of warnings/messages that drivers might receive
from the system and the type of actions that drivers are supposed to undertake.
In condition B, namely, learning by using the multimedia training tool,
participants were trained on the selected system, not only through text
information but also through videos and simulated examples (e.g. sounds)
projected on a PC screen [2]. In essence, the same information provided into
the manual was also provided to the text of the MMT. In addition to these, there
were videos and simulated examples.
In condition C, namely, learning by using the multimedia training tool and
providing hand-on experience, participants were given the additional opportunity
to have a realistic experience on how systems would work in practice, by driving
a driving simulator. Moreover, in this condition they were experiencing situations
where system limitations appear. The simulators used in the three sites were of
the same manufacturer, so exactly the same scenarios have been used in each
site.
The manual and the multimedia training tool were developed in the English
language, as this was the official language of the project, within the framework
of which the present survey was conducted. There were not enough resources
available to have the material translated into the native language of each of the
three countries.
For the evaluation of training methods’ effectiveness, a set of four
questionnaires with multiple choice questions (one set for each system with 5-7
questions per system) was administrated to participants.
220
Tools and methodologies for ITS design and drivers awareness
User ID:
Date:
Group:
_____________________
_____________________
? 1
? 2
? 3
Questionnaire for LDW
Please answer the following questions. Only one answer is correct.
1. When the LDW system is active, steering is:
? Controlled by the system to follow a trace
? Controlled by the system to keep the position in the lane
? Controlled by the system according to the traffic flow
? Controlled by the driver according to the received LDW information
2. In which of the following situations will the LDW warn you properly?
? If the vehicle in the adjacent lane is coming towards you
? If you are drifting to another lane without using your direction indicator
? If you are drifting to another lane, even if you are using your direction
indicator
? If you are drifting to another lane where another vehicle is already
occupying it.
3. Which one of the following road types would be the most appropriate for
LDW?
? Rural roads
? Urban roads
? Highway
? Peri-urban roads (i.e. ring-road)
4. Which preconditions in the road infrastructure should be met in order to get
the LDW to work properly?
? Good road surface
? Good lane markers with a specific range of lane widths
? Open road with good lane markers
? None
5. What should you do if you get an LDW warning?
? Return immediately to your original lane
? Return to your original lane if you did not want to change lane, but keep on
to the next lane if you purposely did it
? Operate your change lane indicator and continue the manoeuvre
? Return to your original lane if you did not want to change lane, but keep on
to the next lane if you purposely did it checking that there is no other car at
the adjacent lane that did not foresee your lane change manoeuvre.
6. Which of the following is not an effect of the LDW?
? Enhance driver comfort
? Enhance traffic safety
? Reduce driver responsibility
? Improve steering performance
Fig.1. Questionnaire used for the Lane Departure Warning system
2.4
Procedure
Participants were seated in a quiet room and were administrated by the
experimenter an entry questionnaire asking general questions about their
characteristics (e.g. age, gender, driving experience, previous experience with
ADAS/IVICS/DSS).
221
Human Centred Design for Intelligent Transport Systems
After a brief description of the scope of the experiment, all participants were
given sufficient time to become familiar with the information provided about the
systems, either by reading the manual or by using the multimedia training tool
for each one of the systems. The participants were asked to read carefully the
manual or to browse themselves all the pages of the MMT reading all the
information included and activating all multimedia content. The experiment
leader asked the participants whether they felt confident that they had
understood well the manual or the MMT. According to their answer, they were
either given more time for the study or they proceeded further. No assistance
was provided by the experimenter in regard to better understanding the context
of the information provided. But since all material was written in English, the
experimenter was however allowed to translate expressions into the native
language of the participant.
In the multimedia training + practice condition, apart from training by using the
multimedia training tool, participants were also provided with a realistic
experience on how systems would work in practice by driving in the driving
simulator. Initially, there was a warm-up scenario, so that the participants were
familiarised with the driving simulator. As two environments were used in the
scenarios, highway and rural road, the participants were driving for 5 minutes in
each environment. Then, there were four categories of scenarios, those related
to ACC, LDW, ABS and the navigation system. The experiment leader was
giving the following instructions to the participants:
Navigation scenarios: Drive following the traffic rules. Try to follow the
navigation system instructions so as to reach destination. Do not overtake.
ABS scenarios: Drive normally without overtaking.
ACC scenarios: Drive in the right lane, without changing lane.
LDW scenarios: Driver in the right lane, imitate that you are drifting towards the
road marker (lane exit) without using indicators.
For each category there were the following scenarios, one showing the normal
functionality of the system and a second showing a limitation of the system:
Navigation scenarios
•
•
Normal functionality
With roadworks
ABS scenarios
•
•
Vehicle braking in good weather with ABS
Vehicle braking in bad weather with ABS
ACC scenarios
•
•
Slow vehicle ahead
Sudden cut in
LDW scenarios
•
•
Good weather
Bad weather
222
Tools and methodologies for ITS design and drivers awareness
2.5
Analysis of results
The analysed measures are the correct answers of the participants to the
questionnaires after the end of each training. We have compared the number of
correct and wrong answers in all questions among group A (learning due to the
use of a manual) and group B (learning due to the use of the MMT). Then, we
have compared the number of correct and wrong answers among group B
(learning due to the use of the MMT) and group C (use of MMT and experience
in a driving simulator), summing up for each system only the questions that are
of relevance for the simulator, namely Q1 and Q6 for ACC, Q2 and Q4 for LDW,
Q3 and Q5 for navigation system, Q1 and Q2 for ABS. The chi2 test was used
for both comparisons.
The graphs below depict the participants’ performance in each question
separately. The reason is that we wanted also to analyse also the
understandability and usefulness of each question. If the majority of people had
failed or respond worse in one question than in the rest questions, this could be
due to the phrasing of the question itself, moreover since the training material
and the questionnaires were not in the participants’ native language.
Country effects were not analysed, as the sample size was very limited, to
acquire meaningful results.
3 Results
The percentages of correct answers per system and per group are given below.
Significant differences were found for the Navigation system (p<0.05, effect
size=0.86), where participants in the Manual group answered more correctly
than participants in the MMT group and for the LDW system (p<0.05, effect
size=0.71), where we note the opposite, participants in the MMT group
answered more correctly than participants in the Manual group.
LDW questions
ACC questions
100%
80%
60%
40%
20%
0%
MMT
Manual
Q1
Q2
Q3
Q4
Q5
100%
80%
60%
40%
20%
0%
Q6
MMT
Manual
Q1
ABS questions
MMT
Manual
Q2
Q3
Q4
Q3
Q4
Q5
Q6
Q7
Navigation questions
100%
80%
60%
40%
20%
0%
Q1
Q2
100%
80%
60%
MMT
40%
20%
0%
Q5
Manual
Q1
Q2
Q3
Q4
Q5
Fig.2. Correctness of answers per system among Groups A and B
223
Human Centred Design for Intelligent Transport Systems
No effect of the driving simulator was found for any system, according to the
analysis of correct answers in the questions which were relevant to the
simulator scenario applied.
Navigation questions
MMT only
100%
80%
60%
40%
20%
0%
Q3
MMT only
Q1
100%
80%
60%
40%
20%
0%
Q2
LDW questions
MMT + simulator
MMT only
MMT + simulator
100%
80%
60%
40%
20%
0%
Q5
ACC questions
Q1
ABS questions
MMT + simulator
Q2
Q6
MMT + simulator
MMT only
100%
80%
60%
40%
20%
0%
Q4
Fig.3. Correctness of answers per system among groups B and C (only relevant
questions)
The participants’ answers to the question if they think that they have learnt to
use the system after the training are given below. The chi2 test was used to
evaluate the distribution of answers in each group. Answers were given in a 5
rate scale, surely no: -2, rather no: -1, neutral: 0, rather yes: +1, surely yes: +2.
There were effects found for the LDW (p<0.01) and for the ABS (p<0.05), where
in both cases participants in the simulator group are more convinced that they
have learnt to use the system than participants in the other two groups. The
figure below have derived after we have transformed the qualitative answers
into numbers and calculated the mean for each group and system.
+2: Surely yes, -2: Surely No
Have you learnt to use the system?
1.60
1.40
1.20
1.00
0.80
0.60
Group C
Group B
Group A
0.40
0.20
0.00
Navigation
ABS
ACC
LDW
Fig.4. Participants’ feelings about learning effectiveness per system among
Groups
The participants answers to the question if they think that they need additional
training before using the system are given below. There were significant effects
found for the LDW system (p<0.01), where participants in the simulator group
224
Tools and methodologies for ITS design and drivers awareness
are more confident that they do not need additional training participants in the
other two groups. Effects were also found for the Navigation system (p<0.05),
where participants in the MMT group are more confident that they do not need
additional training participants in the other two groups.
-2: Surely yes, +2: Surely No
Need for additional training?
1.20
1.00
0.80
0.60
0.40
0.20
0.00
-0.20
-0.40
-0.60
Group C
Group B
Group A
Navigation
ABS
ACC
LDW
Fig.5. Participants’ feelings about need for additional training per system among
Groups
4 Conclusions
Compared to training with only a paper manual, the MMT showed advantages
in the case of only one of the four systems studied, the Lane Departure Warning
system. However, no clear effect of the additional training via the driving
simulator on the correctness of answers was found for any of the systems.
However, people trained with both the MMT and the simulator feel better trained
in the case of LDW and ABS. Moreover, in the case of LDW, people trained with
the MMT and the simulator feel that they need less additional training than the
people trained with the MMT only or with the manual.
These results are rather inconclusive as regards the potential of the MMT and
of the driving simulator to improve acquisition of knowledge by the participants,
as regards the functionalities and limitations of new systems, although positive
effects were found for the MMT in one of the four systems. Further development
of the MMT and of the scenarios employed in the driving simulator, or selection
of different scenarios, showing more clearly the limitations of the system, could
lead to higher impacts on knowledge acquisition. The language issue could
have also played a role in this.
It must be noted that the evaluation of the training effectiveness was based on
number of correct answers in a paper quiz, that is the acquisition of knowledge
by the participants was analysed. Results could be different if evaluation was
based on actual driving assessment in real road or on simulator, namely if we
had analysed the skills acquisition, and if long-term effects could be studied.
Moreover, the sample mainly consisted of people having a university degree, of
whom it can be expected that they can easily learn well by only reading a paper
225
Human Centred Design for Intelligent Transport Systems
manual. Results could be different if more people with lower educational
background were included in the sample.
As a conclusion, we may say that further research is needed, with bigger
sample size coming from various educational backgrounds and with analysis of
actual driving behaviour, in order to clarify if there is a positive impact from the
use of a MMT and of a driving simulator in the training of drivers. Emphasis
should be put on analysing possible negative side impacts of these tools, like
creation of overconfidence, as possibly implied by the participants’ subjective
perceptions in the present study.
5 References
[1] Bekiaris, E., Georgopoulos, P., and Mousadakou, A.: ‘Multimedia training
tool for road safety and traffic behaviour training’. Road Safety training in
high schools (6th Grade), Hellenic Ministry of Transport & Communications,
April 2006.
[2] Simoes, A., and Nikolaou, S.: ‘Deliverable F.2 Inventory of drivers’ needs
for training regarding ITS according to driving tasks affected’. Report of
Task Force F, HUMANIST Network of Excellence, January 2006.
[3] Twisk, D., and Nikolaou, S.: ‘Deliverable F.1 Inventory of ITS functionalities
according to driving task models’. Report of Task Force F, HUMANIST
Network of Excellence, January 2005.
[4] Bekiaris, E., Papakonstantinou, Ch., Stevens, A., Parkes, A., Boverie, S.,
Nilsson, L., Brookhuis, K., Van Wees, K., Wiethoff, M., Damiani, S., Lilli, F.,
Ernst, A., Heino, A., and Widlroither, H.: ‘Deliverable 3/8.1 Compendium of
existing Insurance schemes and Laws, risk analysis of ADA systems and
expected driver behavioural changes. User awareness enhancement,
dissemination report and market Analysis and ADAS marketing strategy’.
ADVISORS Project, March 2001.
[5] Heijer, T., Oei, H.L., Wiethoff, M., Boverie, S., Penttinen, M., Schirokoff, A.,
Kulmala, R., Heinrich, J., Ernst, A.C., Sneek, N., Heeren, H., Stevens, A.,
Bekiaris, E., and Damiani, S.: ‘Deliverable 1/2.1 Problem identification,
User Needs and Inventory of ADAS’. ADVISORS Project, July 2001.
226
SESSION 5: DIVERSITY AND
SPECIFICITY OF ROAD USER
GROUPS
227
Diversity and specificity of road user groups
APPLYING THE RESPONSE CODE OF PRACTICE
FOR EVALUATION OF DRIVER ASSISTANCE
SYSTEMS: HOW THE AGE OF THE DRIVER
INFLUENCES PERCEPTION OF STEERING
TORQUE SIGNALS
Gerrit Schmidt (Volkswagen AG, Germany)
gerrit.schmidt@volkswagen.de
ABSTRACT: This paper describes the effects of additional steering torque as a
haptic signal on the driver-vehicle interaction. Its use for lane departure warning
and lane keeping systems is conceivable, for example. For proper signal design
with regard to the controllability of driver assistance systems required by the
RESPONSE Code of Practice, these signals should be tested on potentially
weaker driver groups. This requirement is taken into account here and therefore
both learner drivers and older drivers were tested in this real vehicle study on
test track. The analyses compare the effect of the signals among these
subpopulations with the effect on middle-aged drivers. A total of 30 drivers were
investigated and metrics of the driver and vehicle reaction were examined. The
results do not indicate that age influences reaction and perception. Neither the
reaction times nor the strength of the reactions differ significantly between the
age groups.
1 Haptic signals and their use for driver assistance systems
The stark increase in information and assistance features in vehicles has
brought with it a constant increase in the number of acoustic and visual signals.
The driver’s haptic channel is used comparatively little for the transfer of
information even though very fast information preparation was proven in many
studies. For example, it has been shown that, in comparison with acoustic
feedback, haptic information (vibration of the handwheel) to support lane
keeping leads to faster correction by the driver when the car is about to leave
the lane [1]. Due to the faster correction, a lower maximum deviation of the
vehicle from the centre of the lane can be observed (figure 1, left).
Figure 1 on the right shows, however, that signals of different modality lead to
the same deviations on average. A closer analysis of the reactions to the
applied steering torque clearly shows the weak points of haptic signals: Some
test persons interpreted the signals incorrectly and responded by steering in the
opposite direction (figure 1, right: Incorrect Strategy). In this case, the
consequence is large lateral deviations [2]. This problem underlines how
important a good ergonomic design of the haptic signal is.
Unfortunately in almost all of the known studies exclusively developed signal
characteristics and scenarios were investigated. This makes it virtually
impossible to make a general statement on design recommendations of haptic
229
Human Centred Design for Intelligent Transport Systems
signals. A systematic investigation of signal characteristics like amplitude or
gradient and of situation features like the current steering torque or the steering
activity in the initial situation has been missing so far. Before using additional
torque for driver assistance systems (e.g. lane departure warning), haptic
signals in the vehicle need to be investigated systematically. In addition to
signal design parameters, situational and usage contexts need to be taken into
consideration in particular.
This paper therefore looks at the effects of directed additional steering torques
in the form of haptic signals in the steering. Unlike vibrations, this type of haptic
feedback contains direction information and, at the same time, shows effects on
the vehicle movements. In the study presented here, the signals were
investigated driving straight ahead. Results on the effect of different driving
manoeuvres on the driver-vehicle interaction are presented in [3].
Fig.1. On the left, the comparison of acoustic (sound) and haptic warning
(torque) modality according to [1]. At the top right, the comparison of acoustic
warning (sound) with two haptic signals (vibration/torque) mod. according to [2].
At the bottom right, distribution of the reaction to pulse-like torque of correct and
incorrect steering strategy of test persons.
2 RESPONSE Code of Practice: Controllability of ADAS
In addition to generally understanding the relationships of vehicle and driver
reactions when additional steering torque is applied, the maximum allowable
interventions should be defined before such signals are used. Therefore it
should be ensured that the driver is not distracted from driving by the
application of such an additional steering torque and is always able to control
the vehicle safely. This central aspect of controllability is covered by the
RESPONSE Code of Practice [4]. This agreement of different OEMs defines a
process for the design and the assessment of driver assistance systems.
Controllability is defined in the Code of Practice as the “likelihood that the driver
can cope with driving situations including ADAS-assisted driving, system limits
and system failures” [4, p5].
230
Diversity and specificity of road user groups
Fig.2. Driving performance-related risk of causing an accident, car accidents for
1000 persons every 1 million kilometres driven, [8; p43].
The Code of Practice proposes alternative approaches to test this aspect. One
suggested possibility is testing at least 20 “naïve” test persons. This suggestion
is taken into account in this investigation and the effects of the torque signals on
normal drivers are looked at. Furthermore potential subpopulations at less or
more risk should be taken into consideration according to the Code of Practice.
Accident analyses [5; 6] show that the risk of causing an accident is above
average, on the one hand, among people over 65 years of age and, on the
other, among young drivers under 25 years of age. Figure 2 shows the risk of
causing an accident related to age. Among older drivers, the reasons include
deterioration in vision, motor functions and the ability to process information
although considerable individual differences can be observed [7; 8; 9]. As a
rule, older drivers are able to compensate for many of these performance
reductions by adjusting their driving style [7, 10]. The reasons for increased
accident responsibility among drivers include a lack of driving experience and a
high level of risk taking [11; 12]. Younger and older drivers have therefore been
considered as driver groups with potential risk in this study.
3 Questions
The study aims to describe the driver’s perception of and reactions to additional
steering torque signals. This should provide design recommendations for haptic
signals that are transmitted via the handwheel. These results are relevant for
the configuration of further steering assistance functions, for example, lane
keeping systems or yaw moment compensation [13].
With regard to guaranteeing the controllability of interventions via additional
steering torque signals, test situations need to be found that allow useful
determination of reasonable interventions and their validation with normal
drivers. One central aspect in this study is the investigation of potentially weaker
driver groups. If it is shown that the applied torque signals lead to other results
among potentially weaker driver groups like new drivers with little driving
experience and elderly drivers than with middle-aged drivers, these groups
should be given special attention in further investigations. To this end,
subpopulations of different ages are confronted with additional steering torque
signals of different amplitudes. The effects on the driver and vehicle are then
231
Human Centred Design for Intelligent Transport Systems
recorded. This study thus makes a contribution to the test design in the
investigation of steering system faults as part of the guidelines from the
RESPONSE Code of Practice [4].
4 Test design
4.1
Driving scenario
In previous internal investigations into the reaction of drivers to steering torque
signals, corners in particular were looked at in addition to the application while
driving straight [13]. These driving scenarios have been further developed for
this investigation. A bottleneck was selected here as a worst case scenario for
driving straight ahead and was simulated using a 2.5m wide track marked with
traffic cones. This bottleneck represents, for example, lanes becoming narrower
at motorway road works and its size is based on the minimum requirements
from road-building guidelines [14]. Since the width of the vehicle used is 1.82m,
the bottleneck only allows very small deviations from the lane guidance. The
driver therefore needs to be able to correct errors very quickly and precisely.
In addition, torque signals on curves of different radii and during the dynamic
manoeuvre “single lane change” were carried out in compliance with the
guidelines from the ISO standard [15]. These measurements are not covered
here. See [3] for the comparison of the driving manoeuvres. The driving
situations were performed in an oval in a fixed order (figure 3). The driving
speed was set at a constant 80km/h using a cruise control system. The task for
the driver was to drive through the four scenarios without hitting the cones or
leaving the lane in the curves. While driving through the scenarios an additional
steering torque was applied. This force resulted in a movement of the
handwheel the driver had to cope with.
Fig.3. Realised driving situations on the test course.
4.2
Variation of additional steering torques
Additional moments with four different amplitudes are set within the different
driving situations. As regards, the amplitude levels, 0<A<B<C applies. The
additional signals with amplitude 0 were integrated in the procedure to provide a
baseline measurement. No additional torque was applied, a measurement was
232
Diversity and specificity of road user groups
simply triggered in the respective situation. All depicted torque signals were
repeated three times and randomly distributed to the right and left. The order
was completely random among the test persons.
The torque signals were held up to approx. 2s after triggering. The reduction is
very slow so that the driver is not confronted again with a fast change in torque.
The test persons are asked for an assessment immediately after completion of
the steering torque signal. The effects of the additional moment were assessed
with the aid of the disturbance assessment scale [16] (figure 4). Five categories
are distinguished on the scale. The test persons made the assessment in a twostage procedure using a scale with a total of 11 stages.
4.3
System setup and investigated sample population
A VW Passat B6 equipped with measuring instruments was used for the test. In
the test vehicle, it was possible to apply additional steering torques using
modified control unit software for the electromechanical power steering. The
investigator accompanied the tests in the front passenger seat and triggered the
additional torque signals invisibly for the test persons during the test drive.
As the evaluation of the torque signals was to be made by normal drivers,
participants were sought who had no experience in the development of driver
assistance systems. 30 test persons, including 22 men and 8 women, took part.
In addition to 16 drivers between 25 and 55 years (m=38.12; sd=5.24), seven
participants between 18 and 25 (m=20; sd=1.92) and seven persons over 55
years (m=66.83; sd=6.43) were investigated. The oldest participant was 78
years old at the time of the test. The average distance driven was for
participants from 25 to 55 years 16063km/year (sd=6875), for under 25 years
17143km/year (sd=4298) and over 55 years 15250km/year (sd=6364).
4.4
Measuring variables and data processing
The recorded physical driving data is filtered (5Hz low pass) and corrected for
offsets. The triggering and the respective vehicle statuses are checked during
data processing. 613 measurements with applied additional steering torques
were used in the analyses. Minimums and maximums of the steering and
vehicle reactions are calculated according to the application of the additional
torque. The maximum handwheel rate after the first maximum of the handwheel
angle is used as a characteristic value for the strength of the driver reaction
(figure 4). As it is not the level, but more the change that is relevant for the
driver, for the yaw rate the span was calculated as the difference between
minimum and maximum within two seconds after the torque signal was
triggered. The calculated characteristic values are combined with the
evaluations. The average values are used in the analyses for the repeated
faults. For all steering inputs and vehicle reactions a smaller value indicates a
better coping of the situation. Due to the experimental design, Split-Plot
ANOVAs with repeated measurements are used for the statistical analysis. The
amplitude of the additional torque is used in the analysis as within subject
factor, the age group as between subject factor. The post-hoc testing of
different factor levels is performed using LSD tests.
233
Human Centred Design for Intelligent Transport Systems
Fig.4. Time pattern of target additional torque, handwheel angle and speed
(left). Disturbance rating scale according to [16] (right).
5 Results
The following results show the effects of different amplitudes of the additional
steering torques. The direct effects of the torque on the steering as well as the
time and strength of the driver reaction are taken into consideration. In terms of
the dynamic reaction of the vehicle to the additional torques and the subsequent
driver entries, the yaw reaction is analysed and set in relation with the
subjective assessments. The analyses compare the different age groups.
Fig.5. First maximum of the handwheel angle (left) and its time (right) per age
level depending on the amplitude of the torque signal.
Figure 5 shows the effects of the additional steering torque on the first
maximum of the handwheel angle. For all amplitudes greater than 0, an
average movement of the handwheel of 3.77deg results due to the torque
application. In the following Split-Plot ANOVAs, only the amplitudes A, B and C
are taken into consideration. Incorporating the baseline amplitude 0 is not useful
since the effect strengths should be illuminated in the different levels. With
regard to the influence of the amplitude of the steering torque, there is a greater
handwheel movement at the amplitude B and C compared with A (p=0.011;
η²=0.149; LSD post hoc Tests: pA-B=0.002; pA-C=0.018; pB-C=0.657). This shows
that up to a certain amplitude, the effects on the handwheel angle rise. At
234
Diversity and specificity of road user groups
amplitudes B and C, the signal is compensated by the driver after the same
handwheel movement for the signals used here. It cannot be presumed that
there are differences between the age groups (p=0.984). Figure 5 (right) shows
the reaction time by the time of the first maximum of the handwheel angle after
application of an additional moment. This maximum marks the time when the
driver is actively intervening to the applied torque. As regards the speed of the
reaction, no difference between the age groups can be found here (p=0.714).
Younger and older drivers react to the additional torque signals applied here on
average equally fast after 431ms. The reaction time of amplitude level A to the
two other levels B and C rises significantly here (p=0.002; η²=0.201; LSD post
hoc Tests: pA-B=0.003; pA-C=0.001 pB-C=0.334). As for the higher amplitudes with
a limited slope it takes more time until the signal has inclined until the desired
amplitude is reached, it can be presumed that the reason for this longer reaction
time is due to this longer inclination and not to slower driver response in these
cases.
The severity of the driver reaction is described using the maximum handwheel
angle rate after the first maximum of the handwheel angle. Figure 6 shows this
maximum of the handwheel angle rate on the left-hand side. The observed
driver reactions are well below 100°/s on average. These values speak for a
quite slow and controlled reaction and support the assumption that the
additional steering torque signals presented here were easy to control by the
driver. The severity of the reactions is not distinguished between the age groups
(p=0.410). At amplitude A, lower values occur compared with B and C (p=0.001;
η²=0.214; LSD post hoc tests: pA-B=0.005; pA-C=0.002; pB-C=0.100). The strength
of the yaw reaction shown on the right in figure 6 behaves in a similar way: Here
too, there is no difference between the age groups (p=0.496). A greater
amplitude leads, however, to greater yaw reactions (p<0.001; η²=0.625; LSD
post hoc tests: pA-B=0.004; pA-C<0.001; pB-C<0.001).
Fig.6 First maximum of the handwheel angle rate after driver intervention (left)
and its maximum yaw rate span (right) per age level and amplitude of the torque
signal.
The subjective ratings of the disturbance effects in figure 7 on the left shows a
significant increase in the ratings with increasing amplitude of the applied
additional torque (p<0.001; η²=0.765; LSD post hoc tests: pA-B<0.001; pAC<0.001; pB-C<0.001). The assessment of the lane deviation is shown on the
right in figure 7. At the greatest amplitude C, the deviation from the centre lane
position is rated as greater compared with the lower amplitudes (p<0.001;
η²=0.384; LSD post hoc tests: pA-B=0.060; pA-C<0.001; pB-C<0.001). There are
no significant age effects (disturbance rating: p=0.239; lane deviation: p=0.240).
235
Human Centred Design for Intelligent Transport Systems
Fig.7. Rating of the effects of the additional torque (left) and the deviation from
the lane centre (right) per age level depending on the amplitude of the torque
signal.
6 Summary
To summarise, it can be stated that no difference between older and younger
drivers can be observed on the level investigated here. However, it should be
mentioned that the volunteers who took part in this study tended to be people
with an affinity for and who enjoy driving. Drivers with performance limitations,
who increasingly face problems on the road, hardly ever volunteer for this kind
of driving study. The recruitment of older weaker drivers poses a dilemma that
seems impossible to solve for the time being. In the selection for this study, it
was important that older participants were taken into consideration.
By investigating younger drivers, we aimed to look into driver experience as a
factor. However, the results here do not confirm that little driving experience
negatively influences the reaction and perception of additional torque signals.
This takes the requirements from RESPONSE into account and includes
potentially weaker driver groups in the investigation. It should be expressly
stated that the aim was not the identification of a sensitive manoeuvre to
differentiate age differences. Moreover the manoeuvre was selected on the
basis of ecological validity: It represents a maximum requirement in road traffic
with regard to the precision of the lateral guidance. Nevertheless, it can be
criticised that these test track scenarios might not cover the full complexity of
situations in real traffic.
In the presented study, both older and younger drivers showed the same
performance as the middle-aged participants in the set driving task and
therefore the results do not indicate that the age is influencing the perception
and reaction to steering torque signals. Not least due to the small sample sizes
of the subgroups in this experiment, further studies with the same setup are
needed for replication.
7 References
[1]
Sato, K., Goto, T., Kubota, Y., Amano, Y., & Fukui, K. (1998). A Study on a
Lane Departure Warning System Using a Steering Torque as a Warning
Signal. Paper presented at the AVEC '98, 14.-18.Sept., Nagoya, Japan.
236
Diversity and specificity of road user groups
[2]
[3]
[4]
[5]
[6]
[7]
[8]
[9]
[10]
[11]
[12]
[13]
[14]
[15]
[16]
Suzuki, K., & Jansson, H. (2003). An analysis of driver's steering behaviour
during auditory or haptic warnings for the designing of lane departure
warning systems. JSAE Review, 24(2003), pp. 65-70.
Schmidt, G. (2007). Wann spürt der Fahrer überhaupt? Der Einfluss des
Fahrmanövers auf die Wahrnehmung von Zusatzlenkmomenten als
haptische Signale im Fahrzeug. 4. VDI-Tagung Fahrer im 21. Jahrhundert
- Human Machine Interface, 14.-15.11., Braunschweig.
PReVENT (2006). RESPONSE 3 - Code of Practice for development,
validation and market introduction of ADAS. Link from 09.07.2007:
http://preventip.org/download/deliverables/RESPONSE3/D11.2/Response3_CoP_v3.0.p
df
Statistisches Bundesamt, 2004. Fachserie 8, Reihe 7. Germany.
Oswald, W. D. (1999). Automobilismus und die "gefährlichen Alten". In G.
Schmidt (Hrsg.). Technik und Gesellschaft. Jahrbuch 10: Automobil und
Automobilismus. Frankfurt: Campus, pp. 183-195.
Tränkle, U. (Hrsg.). Autofahren im Alter. Köln, 1994.
HUK-Verband (Hrsg.) (1994). Fahrzeugsicherheit 90. Analyse von PkwUnfällen. Grundlagen für künftige Forschungsarbeiten. München: HUKVerband.
Sekuler, R. & Sekuler, A. B. (2000).Vision and aging. In: A. Kazdin (Ed.),
Encyclopedia
of
Psychology.
http://www.unet.brandeis.edu/~sekuler/ABSTRACTS/Encyclopedia.html
[Link from 23.02.06].
Hakamies-Blomqvist, L., Raitanen, T. & O’Neill, D. (2002). Driver ageing
does not cause higher
accident rates per km. Transportation Research
Part F, 5, pp. 271- 274.
Boyce, T. E. & Geller, E. S. (2002). An instrumented vehicle assessment of
problem behaviour and driving style: Do younger males really take more
Risk? Accident Analysis and Prevention, 34 (1), pp. 51-64.
Siegrist, S. (Ed.) (1999). Driver training, testing and licensing - towards
theory-based management of young drivers' injury risk in road traffic.
Results of EU-Project GADGET, work package 3 (Bfu-Report 40). Bern:
Schweizerische Beratungsstelle für Unfallverhütung (Bfu).
Schmidt, G., Switkes, J. P., Kiss, M. & Gerdes, J. C. (2007). Auswirkungen
haptischer Signale in der Lenkung auf das Fahrer-Fahrzeug System. In
GZVB (Hrsg.), AAET 2007. Braunschweig: GZVB. 26. - 28. Februar 2007.
Richtlinien für die Sicherung von Arbeitsstellen an Straßen – RSA . Link
vom 09.07.2007: http://www.stvzo.de/seminare/rsa/RSA_Inhalt.htm
ISO 3888-1 (1999). Personenkraftwagen - Prüfgasse für den
Spurwechseltest - Teil 1: Doppelter Fahrspurwechsel. Beuth Verlag.
Neukum, A. und Krüger, H.-P. (2003). Fahrerreaktionen bei
Lenksystemstörungen – Untersuchungsmethodik und Bewertungskriterien.
VDI-Berichte, Reifen-Fahrwerk-Fahrbahn, 1791 , S. pp. 297–318.
237
Diversity and specificity of road user groups
EFFECT OF ADDITIONAL TASKS IN DRIVING
PERFORMANCE: COMPARISON AMONG THREE
GROUPS OF DRIVERS
Marta Pereira (UTL / FMH, Portugal), Houda Hamama (INRETS, France),
Marie-Pierre Bruyas (INRETS, France), Anabela Simões (ISEC, Portugal)
msopereira@gmail.com
ABSTRACT: The growing introduction of new technologies inside vehicles
represents a set of extra information sources that drivers have to manage at the
same time. Their use can interfere with the driving activity and induce
performance decrements. An on-road experiment was conducted to investigate
how drivers reacted to the inclusion of an additional task and how this situation
interfered with guidance and driving activities. Forty eight subjects (elderly,
experienced, novice) participated in the tests. During the course they interacted
with a navigation system and a mobile phone and were their performance levels
evaluated on four parameters: navigation errors, turning indicator activation,
visual behaviour and critical situations management. As a result, phone
conversation did not significantly affect the guidance task. Concerning the
driving, the visual behaviour was less efficient and the activation of the turning
indicator was delayed. Young drivers were least affected by the conversation
while elderly drivers were the most affected in all the parameters.
1 Background
The development of new technologies and their recent incorporation into road
vehicles represents a set of extra sources of information for drivers. This implies
drastic modifications of the main activity of driving especially when the driver
has to manage simultaneous information from different sources. This
information can be linked to the driving activity to assist the driver (such as
information from a navigation system or Adaptive Cruise Control) and on the
other hand, it can also not be directly related with the driving activity and may
enter in the car without any safety device (e.g. mobile phones).
One example of such technology, designed to help the driving task, is the
navigation system. Such systems inform the driver about the itinerary, directing
him or her to reach a specific place. As soon as the driver enters the destination
into the device, the system can select the fastest or shortest route, guiding
him/her in the most efficient manner without becoming lost, frustrated or feeling
bad about not finding the way. Many studies showed the advantage of using
electronic route guidance instead of the old paper map. To Wochinger et al. [1]
the interaction with a route system (turn-by-turn) leads to better driving
performances than the use of a paper map. The system allows for faster mean
speeds on all road types, lower workload ratings and less navigation errors.
Moreover, the large amount of cognitive attention required by paper map use
leads to a higher rate of abrupt braking manoeuvres [1]. Conversely, a turn by
239
Human Centred Design for Intelligent Transport Systems
turn indication presents the relevant information in a simple way. This suggests
that a system providing this kind of instructions, rather than complex route
information, is less distracting to the driver and presents the most usable means
of navigation. Thus the use of a route guidance system is more adequate than
paper map use and can efficiently guide the driver in an unfamiliar area [2].
The growing presence of mobile phones inside the vehicle is also a result of the
emergence and popularity of communication technologies. Their use is risky
and raises the problem of driver distraction. According to Redelmeier and
Tibshirani [3], a mobile phone conversation quadruples the risk of being
involved in a car crash. Laberge et al. [4] also tried to evaluate this risk by
comparing two groups of drivers: users of mobile phone while driving and non
users. Their study shows that the risk is 38% higher during a phone call. The
main question is to determine in which way a mobile phone conversation diverts
the driver’s attention and how it interferes with the driving activity. Several
studies stressed driving performance impairments in terms of reaction times,
decision making, lane keeping [5], and lower driving speed [6]. The quality of
perception and information processing is also affected. Indeed, Recarte and
Nunes [7] [8] and also Harbluk and Noy [9] showed that the added workload
due to a phone conversation increases the driver’s effort of attention. At the
perception level, this is conveyed by a reduction of the visual attention with a
decrease of peripheral stimuli detection like checking the mirror and
speedometer [8]. This decrease was explained as the result of a multitasking
situation induced by the mobile phone conversation. A secondary task, like a
mobile phone conversation, increases the workload, competes with the
attention and, as a consequence, diverts the attention away from the primary
driving task.
Regarding the use of in-vehicle technology, it can be considered that individual
driver’s characteristics are very important and can also induce different types of
interactions, which may have diverse consequences on behaviour. One of the
groups of drivers that studies have been focused on is the “Elderly” due to the
fact that they are an increasingly important cluster. Studies have shown that the
aging process can induce some decrements in driving ability [10]. Among other
examples, some declines are related with the visual, cognitive and motor
abilities; difficulty in discriminating relevant information and longer time required
to process it; a decline in selective attention and attention switching. The elderly
are also characterized by being highly distractible and can also be easily
confused by competing sources of information [11]. However in order to
compensate their perceptual-motor degradation they tend to adapt their driving
behaviour and use residual resources. While driving the elderly try to reduce the
stressful mental load they are experiencing and, as a consequence, they are
more conscious, drive slowly, and attempt to over control their actions [12].
When interacting with in-vehicle information systems older drivers can express
especial difficulties because this interaction may represent a factor of distraction
once it imposes the allocation of cognitive resources to an additional task.
During this period, they could be temporarily unable to react appropriately to an
event or even manage the driving task in complex situations due to the risk of
overloading [10].
240
Diversity and specificity of road user groups
Liu [13] conducted a study to evaluate the interaction with such systems and
tried to check if performances, in terms of reaction time, depended on the
modalities of the displays and on the complexity of the messages. Results
showed longer reaction times and an increase of the navigation errors for the
elderly drivers. They also registered higher performance degradation than
novice drivers during complex or critical situations. Dingus et al. [14] also found
an age effect and they stated that drivers felt an important workload effect
during these situations. The authors explain this tendency by the fact that
elderly drivers suffer from an impairment of their cognitive and perceptive
abilities and also due to the fact that these systems need a considerable
amount of attention to process the information. Due to this gap, elderly drivers
felt unable to process the extra information. Indeed, the performance of an invehicle task needs interpretation and decision making. All of this could lead to
an impairment of the driving activity [15]. Therefore, older drivers require a more
stable and user-friendly road environment and in-vehicle information devices
should be designed in order to avoid an increase of their mental workload.
Unless the systems interfaces and the forecasted interactions are ergonomically
designed, they will overload and confuse the driver, especially the elderly [10].
Another group, with extreme importance in the driving context, is the novice due
to their higher crash rates [16] [17] [18] [19]. This group is characterized by
being more exposed to risk because they drive more often at night [17] [20],
travel at higher speeds and at closer following distances [21] [20] [22].
Furthermore, they are more easily distracted by non-driving events [23], they
tend to overestimate their driving ability in being over confident of their
correction of error [24] [20] [23]. In their literature review, Whelan et al. [25]
showed that the lack of skills of novice drivers (information-processing, selfcalibration, hazard and risk perception and situation awareness) is related to
their crash involvement. Their lower driving experience is frequently associated
with less effective approaches to search visual information and with a poorer
ability to process the perceived information [26]. Due to this lack of experience,
novice drivers have not developed the automatisms in the driving task that allow
for fast switching between tasks [23] [27]. They detect hazards slowly and
perceive them as less risky [19], therefore being unable to anticipate and control
efficiently the vehicle in emergency situations [23]. Furthermore, they may use
some devices in a non-optimal way due to their lack of knowledge and risk
awareness [23].
The impact of additional tasks on young drivers was also analysed by some
authors. Lansdown [28] found driver group differences in the accomplishment of
a secondary task where young drivers took longer to react to a stimulus and
had higher glance frequency towards the additional task interface. In fact, in
multitask situations, novices are not as efficient as experienced drivers for
processing useful information due to their lack of ability to automate cognitive
processes [29] [28]. Indeed, with increasing levels of experience, the driving
activity becomes less demanding and the driver can easily share his/her
attention between the driving activity and the secondary task [28]. It seems that
experienced drivers are less affected by the secondary task given the fact that
they have more ability than novices to share their attention [30]. Moreover,
when a situation is cognitively demanding and when it needs a high level of
information processing like turning left, Gugerty et al. [31] stressed an impact of
241
Human Centred Design for Intelligent Transport Systems
mobile phone use on decision making. This kind of process needs an
integration of all the information presented to the driver in order to anticipate the
future events and other users’ actions.
2 Objectives
Several studies have been conducted in order to investigate the interaction
between the driver and one in-vehicle system. However, the increased number
of new in-vehicle technologies raises the problem of having several not
connected sources of information. Therefore a study was developed in order to
investigate how drivers reacted to the inclusion of an additional task and how
this situation interfered with driving and influenced the guidance task
performance. With more detail, the objective was to analyse the consequence
of a mobile phone conversation in the subject’s behaviour while driving with a
route guidance system. It was important to know how drivers managed both
sources of information originated from the guidance system and the mobile
phone, which information they prioritised, and also what kind of errors occurred
during the driving task.
For this study, it is hypothesised that the mobile phone task will have an
influence in both the guidance and driving tasks. In a more specific way, it is
expected that the mobile phone conversation poses a negative impact, inducing
to a strong decrease in the guidance performance and also to a significant
decrement in the driving activity. Concerning drivers’ specificities, it is believed
that elderly and young drivers will be the groups that perform worse due to the
effects of the mobile phone conversation. For the elderly, this result is expected
due to their higher difficulty to face the more complex situations, especially
considering the simultaneous processing of information from different sources.
For young drivers, due to the fact of having less automatic processes, the
introduction of an additional task will increase their mental workload and
degrade the guidance and the driving task. In spite of the expected inferior
results for those two groups of drivers in the presence of the mobile phone, it is
hypothesised that these declines will be even more evident for the elderly
group.
3 Method
Forty eight drivers took part in the experiment. They were distributed in three
groups of 16 participants, depending on their age and driving experience:
elderly, reference and young drivers. Elderly participants were aged from 62 to
78 years (mean age=69.4; SD=3.9), reference subjects from 34 to 47 years
(mean age=39.6; SD=3.9) and novice drivers from 18 to 21 years (mean
age=19.5; SD=1). The elderly and reference subjects had possessed their
driving licences for at least five years and have driven in excess of 10000 km.
Conversely, novice drivers had less than 2 years of driving experience. An
equal number of men and women were represented in each driver group.
All drivers had their own mobile phone: in average, elderly drivers possessed it
for almost 5 years (m=4,63; SD=3,49), reference drivers for 7 years (SD=2,19)
and novice drivers for 5 years (SD=1.26). All but four elderly subjects stated
242
Diversity and specificity of road user groups
that they had already used their mobile phone while driving. The majority of
elderly drivers admitted to having phone conversations a few times per month
while driving. Comparatively, most reference group participants declared having
a phone conversation at least several times per week while the majority of
young drivers stated to use the mobile phone while driving at least once per
day. From the 48 participants only two indicated that they possessed a
navigation system, one belonging to the elderly and other to the reference
group.
This on-road experiment took place near Lyon (France) with the INRETS
vehicle (Institut National de Recherche sur les Transports et leur Sécurité)
equipped with sensors that registered the dynamic data of the car. Five mini
video-cameras allowed for capturing images from the road environment and the
driver and also from the in-vehicle technologies placed in the car. For safety
reasons, the car was equipped with another set of pedals in front of the
passenger seat and a driving instructor was always seated besides the driver.
A guidance system (Carminat) was fitted as standard to the vehicle and was
located on the top of the dashboard to the right of the steering wheel. It
displayed schematic guidance instructions throughout arrows (turn-by-turn
system) and also transmitted audio instructions some meters prior to
intersections.
The recruited participants were asked to drive a predefined course, guided by
the instructions of the guidance system and to interact at some specified
periods with a mobile phone. Twenty “target intersections” were selected from
the course in order to be studied with further detail: 8 turns to the right and 12
turns to the left. To be considered as a “target”, intersections had to allow the
driver to make all the decisions as freely as possible; meaning the driver had to
decide when to turn and the decision not influenced by traffic lights or other
types of vertical signs.
Subjects only drove the course once and for that reason it was designed to
have an equal number of situations in each experiment condition: No Phone
and Phone. In the No Phone condition, participants were asked to drive the car
with the help of the route guidance system. No other in-vehicle systems were
connected nor were other tasks asked of the participants to perform. In the
Phone condition subjects had to drive the car with the help of the route
guidance system and also conduct a mobile phone conversation. Participants
interacted with a bluetooth hands-free phone connected to the audio system of
the vehicle. Two buttons in the centre of the dashboard allowed them to answer
the call (left green button) and to hang up (right red button). During the course
half of the target intersections were performed without phone (4 right turns and
6 left turns) and the other half with. The system conditions were balanced over
the intersections so that half of the subjects in each group performed a set of
target intersections in one condition and the other half in the other condition.
The mobile phone task was compiled of a series of sentences sent by a
researcher located at INRETS. The participants had to listen to each sentence,
repeat it and then answer “Yes” if it was sensible, and “No” if it was not. For
example: “Usually, bicycles are bigger than cars” to which subjects had to
repeat and then answer “No”. The rhythm with which sentences were given
243
Human Centred Design for Intelligent Transport Systems
depended on the pace of the driver to answer the preceding one. This mobile
phone task was based on the “Decision part” of the Working Memory Span Test
[32] [33]. A repetition part was added to the test in order to ensure that the
correct sentence was heard and to better evaluate the accuracy of the answer.
All sentences contained 11 or 12 syllables and took an equivalent duration of
time to be said. The test contained 50% of sensible phrases and 50% of
nonsense sentences. The order in which they were presented was randomly
selected by computer utilising a visual basic program made specifically for
ordering the sentences.
At the beginning of the experiment, the main objectives of the study were
explained to the participants. They were then submitted to visual and audio
tests to ensure they had no related problems that could interfere with the
experiment and compromise the results. Subsequently, they filled a
questionnaire with personal data and also to investigate their opinion and
knowledge about new in-vehicle technologies, especially mobile phones and
guidance systems. Before the experiment itself, the guidance system
functioning as well as the details of the mobile phone task were explained to the
participants. A training period of at least 15 minutes allowed them to get familiar
with the car, the guidance system and the phone task. The possibility to extend
this familiarization period was given to the driver if needed. The experimental
test lasted from 30 to 45 minutes depending on traffic conditions, navigation
errors and also on the speed of each driver. During the test, an observational
table was filled in by a researcher in order to register aspects regarding
navigation, turning indicator signs, driving behaviour, trajectory of the vehicle,
interaction with the other road users, respect of road signs, visual behaviour
towards the more important areas for collecting visual information from the road
environment (such as rear mirrors and the intersections itself). Such a table was
based on Risser work [34]. At the end of the experiment, participants filled a
questionnaire to investigate the attentional demand according to the
experimental conditions (No Phone and Phone).
In spite of several parameters being measured in this experiment (the
performance on the primary task, the secondary task and also self-reported
measures) only the analysis from the navigation errors and relevant driving
errors will be presented in this paper.
4 Results
The variables presented in this paper were only collected at the target
intersections because, just in half of those junctions was the mobile phone
condition introduced. Due to navigation errors that occurred during the
experiment, some participants made small deviations from the original
predefined course, resulting in different numbers of target intersections. The
allowance for such errors was justified by the fact that drivers should manage
their course by themselves in order to find the path and the researcher not
allowed to interfere with their decisions. As a consequence, to compare the
results, percentages of errors have been computed in each condition. Based on
these percentages, results were compared using a statistical test of proportions
that was applied in each case (Fisher’s test). A significance threshold of 0.05
244
Diversity and specificity of road user groups
was accepted (p<5%). The experimental design included one between-subjects
factor: the driver groups (elderly, reference and novice groups) and one withinsubjects factor: the driving conditions (Phone and No Phone conditions). Only
the statistically significant results are presented.
4.1
Navigation Errors
A navigation error can be considered as a discrepancy between the information
sent by the system and the action carried out by the driver. The frequency of
such errors was counted and results concluded that from the 843 driven
intersections, 42 resulted in navigation errors (approximately 5%). Data showed
a higher but non significant percentage of errors in the Phone condition (5.91%)
compared with the No Phone situation (4.12%). Similar results were observed
for each driver group, where the Phone condition produced a higher percentage
of error. Comparing the proportion of errors for the driver groups, elderly drivers
made more navigation errors in both Phone and No Phone conditions.
Reference drivers were the ones that made fewer errors and novice drivers
positioned themselves in an intermediate position, as shown in Figure 1. It is
important to point out that, while in the No Phone situation the differences
between driver groups are not considerable, for the Phone condition the
discrepancies are more pronounced, especially between elderly and reference
drivers. Although these differences are not statistically significant (p=0.07
between elderly and reference for the Phone situation) this shows a tendency
for elderly drivers to make more navigation errors while at the phone.
Navigation Errors
% of Errors.
25%
20%
15%
No Phone
9.4
10%
5%
4.9
Phone
3.3
3.6
4.2 5.0
0%
Elderly
Reference
Young
Age Groups
Fig.1. Percentage of navigation errors for each driver group
4.2
Turning Indicators
The number of errors for the turning indicators activation was counted with the
help of an observational table and afterwards validated using the video
recording of the test. This analysis was divided into two major errors: one
characterized by the lack of turning indicator sign (“omission” error), and the
other considered when the turning indicator was activated too late, i.e., a par of
meters or less from beginning to turn the steering wheel (“timing” error).
By comparing both mobile phone conditions it can be seen that in the Phone
condition a higher percentage of “omissions” and “timing” errors are observed. It
can also be observed that, in this condition, “timing” errors occurred more often
245
Human Centred Design for Intelligent Transport Systems
than “omission” errors. In spite of the dissimilarity between both mobile phone
conditions not being significant for the “omissions”, for the “timing” errors this
difference is statistically significant. Therefore, it can be stated that the Phone
condition led to a significantly higher percentage of late activations of the
turning indicator sign (p=0.01). When the comparison between groups is made
(Figure 2 and 3), it can be seen that elderly drivers are more likely to forget to
activate the turning indicator signal while novice drivers have the smallest
“omission” errors percentage. However, the difference is not statistically
significant within driver group or between them. The only statistically significant
difference occurred for the “timing” errors within the elderly group. Those
participants made higher percentage of errors in the Phone condition than in the
No Phone situation (elderly drivers p= 0.01), meaning that elderly drivers were
the ones significantly more affected by the addition of the phone conversation.
This performance decrement was characterized by a latter activation of the
turning indicators.
"Timing" Errors for Turning Indicators
25%
25%
20%
20%
15%
10%
No Phone
10.0
5.7
5%
Phone
4.6 6.4
2.0
3.3
% of E rrors.
% of E rrors.
"Omission" Errors for Turning Indicators
19.1
15.1
15%
10%
14.9
No Phone
11.1
7.4
5.7
Phone
5%
0%
0%
Elderly
Reference
Elderly
Young
Reference
Young
Age Groups
Age Group
Fig.2. and 3. Percentage of “omission” and “timing” errors for each driver group
4.3
Visual Behaviour
Similar to the turning indicator sign, the number of errors in visual behaviour
was coded through the observational table. This register was also made during
the experiment and afterwards validated utilising the video recording of the test.
The analysis of the visual behaviour was made based on the assumption that
drivers have to attend to specific areas of the road environment in order to
capture the crucial information to perform a turn. Whenever a driver does not
look to these areas, an error was marked down in the table.
From the analysis of the results it is possible to verify that, in a general way, a
higher percentage of “visual behaviour errors” were committed in the Phone
condition. When all subjects are taken into consideration this difference is
statistically significant (p=0.00). This result is also confirmed for each of the
participants’ group where higher percentages of error are observed when a
mobile phone conversation is conducted (Figure 4). However, when ages are
taken into account, only the elderly and the reference participants reveal a
significant difference of errors between mobile phone situations (elderly
p=0.015; reference p=0.00). A comparison between driver groups shows that
elderly drivers made more “visual behaviour errors” in both mobile phone
conditions. Reference drivers made fewer errors in the No Phone situation and
their younger counterparts had a lower percentage of error in the Phone
condition. However, those differences are not significant.
246
Diversity and specificity of road user groups
Visual Behaviour Errors
% of Errors.
25%
22.4
20%
15%
17.3
14.1
10.5
10%
6.6
No Phone
Phone
4.1
5%
0%
Elderly
Reference
Young
Age Group
Fig.4. Percentage of “visual behaviour errors” in every age group
4.4
Near Crash Situations
Near crash situations represent the moments where the driver placed himself
into a dangerous situation and the instructor, travelling beside him/her, had to
intervene. These interventions aimed at avoiding a critical situation or, in the
worse case, a crash between the experimental subject and other road users.
The instructor could intervene by means of the double pedals installed in front
passenger side of the vehicle and also through the direct control of the steering
wheel. Such situations occurred during the present experiment. Video recording
analysis showed that all the 5 near crash situations registered concerned
elderly drivers Moreover, all but one of these critical events coincided with the
usage of the mobile phone. When considering these proportions, the near crash
situations in the No Phone condition represents less than 1%, while in the
Phone situation represents 3.13%. This suggests that for the elderly drivers,
conducting a mobile phone conversation highly increases their probability of
near crash or crash situations.
5 Discussion and conclusion
Regarding the navigation errors, when all driver groups are considered, the
analysis of this data revealed a slight tendency of an increasing number of
navigation errors in the Phone condition. However, contrarily to what was
expected, the Phone situation did not reveal a statistically significantly higher
number of navigation errors. This may be due to the novelty of the guidance
task and to the participants’ inexperience of interacting with this type of system,
leading to numerous navigation errors being made even when drivers were not
having a mobile phone conversation. However, this could also be related to the
imperfect accuracy of the distance given by the navigation system, which was
not always accurate when informing about the number of meters for the next
turn (higher speeds also decreased the distance precision). This lack of
distance precision could have lead drivers to a higher percentage of errors,
even when they were not performing the secondary task. Additionally, it should
be considered that the nature and difficulty of the mobile phone conversation
did not lead to a sufficient involvement of the experiment participants. This
supports the statement saying that, if drivers are not really involved in the
conversation, no visible behaviour impairments can be seen [35] [36]. On
247
Human Centred Design for Intelligent Transport Systems
revealing that the difference between mobile phone conditions was not
significant, it cannot be established that the number of errors is directly
attributed to the mobile phone conversation. For this experiment, the mobile
phone conversation did not lead to higher percentage of navigation errors. In
considering the driver groups, elderly drivers made more errors, especially in
the Phone condition, where the difference between them and the reference
drivers is close to being statistically significant. Thus the elderly show a slight
tendency to make more guidance mistakes than reference drivers in a multitask
situation.
In considering the turning indicator sign, there was no significant difference for
the “omission” errors independently of the mobile phone condition or the driver
group. This means that in this experiment, the mobile phone did not produce an
important effect by inducing drivers to forget to activate the turning indicators.
On the other hand, it can be stated that when all subjects are taken into
account, a significantly higher number of late activations of the turning indicator
were made coinciding with the phone conversation task. In fact, the detailed
analyses of the age groups revealed that elderly drivers were the only group
that had higher rates of “timing” errors while using the phone. Neither the
reference drivers nor the young drivers drastically changed their “turning
indicator” behaviour in the presence of the phone conversation. The data
highlights unequal results between age groups and can reveal different ways to
process the information. According to Dingus et al. [14] elderly drivers may
experience higher difficulties while managing several sources of information and
this can lead to a longer information processing. In the present experiment, this
higher time to activate the turning indicator sign by the elderly can be
considered as the expression of this slowing process.
In the visual behaviour analysis, it is important to highlight that the mobile phone
conversation significantly influenced the visual behaviour increasing the
percentage of visual errors in this situation. This would suggest that, in the
presence of that secondary task, participants more frequently forgot to check
some important areas of the environment. The significance of this finding is that
dangerous situations could arise as a result of a driver receiving and perceiving
less information from the road reducing his/her reactions to an unforeseen
event. When age is taken into account, elderly and reference participants
committed a significantly higher percentage of errors while using the phone,
meaning that the cognitive activity of these two groups of drivers was affected
by the presence of the mobile phone conversation. The young driver also made
more errors during the phone conversation, but the difference was not
significant. According to the literature and supporting the hypothesis in this
study, elderly drivers had visual behaviour decrements in the presence of the
additional task. Contrarily, the moderate impact of the secondary task on the
visual behaviour of the younger drivers’ performance did not confirm the results
of other studies. This can be justified by the assumption that this group of
drivers is more used to interacting with new technologies, coping better with this
multitask situation. This is also supported by a higher reported use of mobile
phones by this sample as registered in the questionnaire completed prior to the
experiment. Using a mobile phone in the car is more familiar to them and the
effect is registered as less important on their driving behaviour. This lack of
effects from the additional task registered in the young drivers could also be due
248
Diversity and specificity of road user groups
to the presence of the driving instructor whose presence could have caused a
feeling of being judged. Additionally, the fact of their relative inexperience to
driving and probably not having much confidence in their abilities could induce
them to pay extra attention to their driving avoiding making mistakes.
The decrements in performance of elderly drivers are once again highlighted
when the results of the near crash situations are analysed. The fact that there is
a higher percentage of near crashes for elderly drivers in the Phone condition
reveals that they have more difficulty managing guidance and mobile phone
tasks at the same time while driving. This assumption supports literature results
that indicate older drivers have more difficulties in processing extra information
coinciding with some performance decrement. It should be highlighted that the
near crash events during the experiment could have become real crashes
without the control of the driving instructor.
The conclusions presented in this paper represent a first look to the results
withdrawn from the performed tests. An additional and deeper analysis of the
other parameters recorded during this experiment will expand conclusions on
this topic and allow for an increased knowledge about the influence of additional
and competitive sources of information on the performance of the driving task.
As a conclusion, and contrarily to what was hypothesised, the additional task of
the mobile phone did not have a crucial impact on the guidance task and the
percentage of navigation errors was not significantly different between mobile
phone conditions. However, as expected, the influencing factor was more
evident in the driving task because, in an overall review, participants activated
the turning indicator later while at the phone. An effect was also found in visual
behaviour, where drivers paid less attention to the important areas of the road
environment to perform an intersection. As hypothesised, when ages are taken
into consideration, it is the elderly whose performance levels change the most in
relation to the late activation of the turning indicator and also for flaws in
checking the road environment. Like the elderly participants, reference drivers
also performed significantly worse in terms of visual behaviour. However, in
opposition to the hypothesis, younger subjects did not seem significantly
affected by the mobile phone conversation, at least when considering the
measures used in this experiment.
6 References
[1]
[2]
[3]
Wochinger, K. and Bohem-Davis, D.: ‘Navigational preference and driver
acceptance of advanced traveller information systems’, in Noy Y. I. (Ed):
‘Ergonomics and safety of intelligent driver interfaces’ (LEA, 1997)
Dingus, T., McGehee, D., Hulse, M., Jahns, S. and Manakkal, N.: ‘Travtrek
Evaluation Task C3 – Camera Car Study.’ (McLean, 1995, Report Nº
FHWA-RD-94-076), in Young, K., Regan, M. and Hammer, M.: ‘Driver
Distraction: A Review of the Literature.’ (Monash University, 2003)
Redelmeier, D. A. and Tibshirani, R. J.: ‘Association between cellulartelephone calls and motor vehicle collisions’, N. Engl. J. Med., 1997, 13;
336(7), pp.453-458.
249
Human Centred Design for Intelligent Transport Systems
[4]
[5]
[6]
[7]
[8]
[9]
[10]
[11]
[12]
[13]
[14]
[15]
[16]
[17]
[18]
[19]
Laberge-Nadeau, C., Maag, U., Bellavance, F., Lapierre, S.D., Desjardin,s
D. and Messier, S.: ‘Wireless telephones and the risk of road crashes.’
Accid. Anal. Prev., 2003, 35, (5), pp. 649-60
Reed, M. P. and Green, P.A.: ‘Comparison of driving performance on-road
and in a low-cost simulator using a concurrent telephone dialling task.’,
Ergonomics, 1999, 42, (8), pp. 1015-1037
Young, K., Regan, M. and Hammer, M.: ‘Driver distraction: a review of the
literature’ (Monash University; 2003, Report No.: 206)
Recarte, M.A. and Nunes, L.M.: ‘Effects of verbal and spatial-imagery
tasks on eye fixations while driving’, J. Exp. Psychol. Appl, 2000, 6, (1),
pp.31-43
Recarte, M.A. and Nunes, L.M.: ‘Mental workload while driving: effects on
visual search, discrimination, and decision making’, J. Exp. Psychol. Appl.,
2003, 9, (2), pp. 119-137.
Harbluk, J. L. and Noy, Y.I.: ‘L'impact de la distraction cognitive sur le
comportement visuel du conducteur visuel du conducteur et sur le contrôle
du véhicule’(Transports Canada; 2002, Report No.: TP3#13889F)
Simões, A.: ‘Elderly drivers and IVIS’, in Brusque, C. (Ed.): ‘The Influence
of In-Vehicle Information Systems on driver behaviour and road safety’
(INRTES, 2007) pp. 59-74
Holland, C. A.: ‘Older Drivers: A Literature Review.’ (Department for
Transportation, 2001, Report No. 25)
Eby, D., Shope, J.T., Molnar, L.J., Vivoda, J.M. and Fordyce, T.A.:
‘Improvement of older driver safety through self evaluation: The
Development of a Self-Evaluation Instrument.’(UMTRI, 2000)
Liu, Y-C.: ‘Comparative study of the effects of auditory, visual and
multimodality displays on drivers performance in advanced traveller
information systems’,Ergon., 2001, 44,(4),pp.425-442.
Dingus, T.A., Hulse, M.C., Mollenhauer, M.A., Fleischman, R.N.,
McGehee, D.V. and Manakkal, N.: ‘Effects of age, system experience, and
navigation technique on driving with an advanced traveler information
system’, Hum. Factors, 1997, 39, (2), pp.177-199.
Blanco, M., Biever, W., Gallagher, J. and Dingus, T.: ‘The impact of
secondary task cognitive processing demand on driving performance
Accid. Analys. Preven. ,2006, 38, (5), pp. 895-906
Drummond, A.: ‘An overview of novice driver performance issues’ (Monash
University, 1989).
Williams, A.F.: ‘Teenage drivers: patterns of risk’, J. Safety Res., 2003,,34,
(1), pp.5-15.
McCartt, A.T., Shabanova, V. I. and Leaf, W.A.: ‘Driving experience,
crashes and traffic citations of teenage beginning drivers’, Accid. Anal.
Prev., 2003, 35, (3), pp. 311-320
Clarke, D.D., Ward, P. and Truman, W.: ‘Voluntary risk taking and skill
deficits in young driver accidents in the UK’, Accid. Anal. Prev., 2005, 37,
(3), pp.523-529
250
Diversity and specificity of road user groups
[20] Gregersen, N. P., and Falkmer, T.:’In-vehicle support system and young,
novice drivers’, in Dorn, L. (Ed.): ‘Driver Behaviour and Training’ (Ashgate,
2003)
[21] Simons-Morton, B., Lerner, N. and Singer, J.: ‘The observed effects of
teenage passengers on the risky driving behavior of teenage drivers.’,
Accid. Anal. Prev., 2005, 37, (6), pp. 973-982.
[22] Boyce, T. E. and Geller, E.S.: ‘An instrumented vehicle assessment of
problem behavior and driving style:do younger males really take more
risks?’, Accid. Anal. Prev., 2002,34,(1),pp.51-64.
[23] Lerner, N. D.: ‘Young drivers and highway design and operations: findings
and recommended research.’ (Research Boulevard, 2001,No. DTFH61-98C-00063)
[24] Brown, I. D.: ‘Exposure and experience are a confounded nuisance in
research on driver behaviour.’, Accident Analysis and Prevention, 1982;14,
(5), pp. 345-352, in Clarke, D.D., Ward, P. and Truman, W.: ‘Voluntary risk
taking and skill deficits in young driver accidents in the UK, Accid. Anal.
Prev., 2005, 37, (3), pp. 523-529
[25] Whelan, M., Senserrick, T., Groeger, J., Triggs, T. and Hosking, S.:
‘Learner driver experience project’(Monash University; 2004, Report No:
221)
[26] McDonald, W. A.: ‘Young driver research program. A review of information
on young driver performance characteristics and capacities’ (Monash
University, 1994)
[27] Heinrich, H.C.: ‘Behavioural changes in the context of traffic safety.’,
IATSS Research, 1990;14, (1), pp.85-8, in McDonald, W. A.: ‘Young driver
research program. A review of information on young driver performance
characteristics and capacities’ (Monash University, 1994)
[28] Lansdown, T. C.: ‘Individual differences during driver secondary task
performance: verbal protocol and visual allocation findings.’, Accid. Anal.
Prev., 2002, 34, (5), pp. 655-662
[29] Summala, H., Lamble, D. and Laakso, M.: ‘Driving experience and
perception of the lead car's braking when looking at in-car targets’ Acc..
Anal. Prev., 1998, 30,(4), pp. 401-407.
[30] Wikman, A.S., Nieminen, T. and Summala, H.: ‘Driving experience and
time-sharing during in-car tasks on roads of different width.’ Ergonomics,
1998; 41, pp. 358-372.
[31] Gugerty, L., Rakauskas, M. and Brooks, J.:‘ Effects of remote and inperson verbal interactions on verbalization rates and attention to dynamic
spatial scenes’,Accid. Anal. Prev.,2004,36,(6), pp. 1029-1043
[32] Baddeley, A., Logie, R., Nimmo-Smith, I. and Brereton, N.: ‘Components of
fluent reading.’, Journal of Memory & Language, 1985, 24, pp. 119-131.
[33] Alm, H. and Nilsson, L.: ‘The effects of a mobile telephone task on driver
behaviour in a car following situation.’, Accid. Anal. Prev., 1995, 27, (5),
pp. 707-715
[34] Risser, R.: Behavior in traffic conflict situations. Accid. Anal. Prev., 1985,
17, (2), pp.179-197
251
Human Centred Design for Intelligent Transport Systems
[35] Patten, C., Kircher, A., Ostlund, J. and Nilsson, L.:’ Using mobile
telephones: cognitive workload and attention resource allocation.’ Accid
Anal Prev., 2004, 36,(3),pp. 341-350
[36] McCarley, J., Vais, M., Pringle, H., Kramer, A., Irwin, D. and Strayer,
D.:’Conversation disrupts change detection in complex traffic scenes.’ Hum
Factors., 2004,46,(3),pp.424-436
252
Diversity and specificity of road user groups
WHAT TOOLS ARE NEEDED TO DEVELOP SAFE
AND JOYFUL CYCLING FOR SENIOR CITIZENS
Lars Leden (VTT, Finland)
Lars.Leden@vtt.fi
ABSTRACT: Demographic changes show that the absolute number and portion
of the population in Europe that can be categorized as older or very old will
continue to grow over the next several years. One aim should be to keep them
active and healthy for as long a time as possible. Exercise, for example cycling,
plays an important role in this context but data shows that the elderly bicyclists
are overrepresented in crashes when compared with their exposure to traffic.
Senior cyclists’ needs and preferences should be a base for developing a safe
and joyful cycling environment. This project uses literature reviews, in-depth
crash data analysis, questionnaires with senior cyclists, questionnaires with
experts, and an expert workshop to identify potential ITS applications for
improving elderly bicycling. The last tool (the expert workshop) included two
group discussions structured according to two philosophically different models:
The Diamond model and The Multiple comfort model.
1 Background and purpose
Demographic changes show that the portion of the population in Europe that
can be categorized as older or very old will continue to grow over the next
several years. The absolute number of older people will also continue to grow
and since there will be more old people, one aim should be to keep them active
and healthy for as long a time as possible. Exercise, for example cycling, plays
an important role in this context – it supports us to stay healthy in all phases of
our lives. The health effects of cycling are well documented [1, 2, 3]. Bicycling is
possible almost without any limitation of age, so bicycling is an ideal way to stay
active at an older age. Apart from the advantages for our physical constitution,
cycling could increase mobility at an older age. The bicycle could become an
ideal means of transportation for many senior citizens, in order to fulfill their
individual needs of mobility, and to stay active and mobile at an older age
provided that bicycling is safe. But data shows that the elderly bicyclists are
overrepresented in crashes when compared with their exposure to traffic [4].
Maring and Schagen [5] support the findings. Therefore, measures are needed.
The following five tools for idea generation were applied for identifying user’s
needs and developing countermeasures for safe and joyful cycling for senior
citizens:
1.
2.
3.
4.
5.
literature review,
in-depth crash data analysis,
questionnaires with senior cyclists,
questionnaires with experts,
an expert workshop with group discussions structured according to two
different models.
253
Human Centred Design for Intelligent Transport Systems
The outcome of each tool is described below. However, the result of the
literature review is presented as a part of the description of the outcomes of the
four other tools. At the end of the paper the outcomes of the five tools are
compared and discussed. For a more detailed description of the results, see [6].
2 An analysis of Finnish in-depth crash data
A set of hypotheses was tested on Finnish in-depth crash data (VALT) to find
out reasons behind the higher risks for senior cyclists. The analysis supports
the following hypotheses:
•
Elderly bicyclists are significantly (p=0.0012) more involved in crashes
when intending to turn left compared to other age groups. 22% of
elderly in fatal crashes intend to turn left compared to 8% for adults and
14% for children. Goldenbeld [7] found similar results, that elderly
bicyclists often have problems at intersections and especially when
turning left.
•
As expected, elderly bicyclists are significantly more often impaired by
bad sight (p=3.52E-05) and/or bad hearing (p=3.52E-05) as well as
being impaired from taking medication (p=7.89E-08) in crashes
compared to other age groups.
•
Elderly bicyclists are less often in a hurry (5%) in crashes compared to
other age groups (11%). Of the bicyclists that were fatally injured in
1995-2001, there were a higher percentage of children (18%) that were
in a hurry than among other age groups (6%). Differences were not
significant.
Somewhat unexpectedly, it was found that:
•
Elderly bicyclists obey traffic rules no more and no less than other age
groups. However, non-elderly adult bicyclists are significantly more
often (p=0.00024) affected by alcohol (50% proven impaired) than
elderly bicyclists (9%).
•
In darkness (incl. dawn and dusk), non-elderly adult bicyclists are
significantly (p=4.1E-10) more often involved in crashes (37%) than
elderly (11%).
•
There is no significant difference between age groups’ bicycle front light
and reflector use, and the footbrake on elderly’s bicycles is not less
often in good working order compared to other age groups’ bicycles.
•
Child bicyclists are significantly (p=0.00035) more often involved in fatal
crashes outside built-up areas (56%) than elderly (39%) and other adult
bicyclists (30%).
•
Elderly bicyclists are not over-involved in crashes where the road
surface is in disrepair.
•
Elderly bicyclists are not significantly more involved in fatal crashes on
hilly streets than other bicyclists.
254
Diversity and specificity of road user groups
•
Adult bicyclists are significantly (p=5,85E-06) more often involved in
single-vehicle crashes compared to other age groups. For crashes
involving other pedestrians, other bicyclists or mopeds there is no
significant difference between age groups.
3 A questionnaire to senior cyclists in Sweden
Interviews with 31 bicyclists (15 men and 16 women), all members of the
Cycling Promotion in Sweden (Cykelfrämjandet), were done as a pilot project to
test and finalize a questionnaire about needs and safety of elderly bicyclists,
see [8]. To gather more extensive knowledge about elderly bicyclists needs, the
questionnaire was sent to a sample of more than 500 members of the Cycling
Promotion in Sweden (Cykelfrämjandet) in June 2007. The sample was
stratified to get a better balance between regions (North, Middle and South of
Sweden) and age groups (65-74, 75-84 and 85-89). Altogether, 351 answers
were received from members 65+, corresponding to a response frequency of
61%. The answer frequency decreased with increasing age and was 61% in
average. 40% of the respondents were female. Seven respondents (2%) were
85+. The oldest one was 89 years.
When interpreting the results below it should be remembered that the
respondents are members of the Cycling Promotion in Sweden and have more
experience in cycling and matters related to cycling than people in general in
Sweden and therefore not representative for all bicyclists of that age. However
having experienced respondents can of course be an advantage when
gathering background information to be used to develop a strategy and
measures to obtain safe and joyful cycling for senior citizens. They are probably
also healthier. The share that finds bad hearing a safety problem is small, only
9%, but increases somewhat after age 75. Below follow results.
The foremost reason that elderly ride bicycles is to get exercise, which 94% of
the respondents state as a reason. Other often stated reasons are: because it is
joyful (84%), because it gives freedom (73%), because it is easy (72%) and
because it is easy to park (66%). Note that on this and several other questions
respondents are allowed to give several answers. The foremost reason that the
elderly leave their bikes at home and use another means of transportation is
bad road conditions during the winter (which also is the reason that so many do
not bike at all during the winter): slipperiness (81%), bad snow removal (79%)
and snowfall (77%). Temperatures below zero Celsius restrain about half of the
elderly from cycling.
Also, long distances are a reason that elderly choose not to use a bike. Some
leave their bikes at home when the distance in one direction is more than 6-10
kilometers. Two thirds (65%) of the respondents do not like biking if the (oneway) distance is above 15 kilometers. Almost half of the respondents state that
their bike usage would increase if there was a possibility to bring the bike onto
busses and trains. This view is especially common (57%) among the youngest
group (65-69). The most common comment is that such a possibility would
facilitate going on bike holidays or longer bike excursions or to use the bike at
the destination.
255
Human Centred Design for Intelligent Transport Systems
The most commonly used equipment is lights, which are used by 81% of the
respondents. Most common are battery-powered lights followed by traditional
dynamo-operated ones where the generator touches the tire. Some
respondents have a dynamo in the hub. The second most common equipment
is a helmet, which is used by 80% of the elderly. The remaining fifth does not
own one. The helmet use is lower in Southern regions. About two thirds of the
respondents use a bicycle-bag or basket and reflectors. The use of reflectors
increases with the size of the municipality. Contrary, reflective vests are used
only by 17% of the respondents, but in rural areas the usage is close to 50%.
Rear-view mirrors are used by a few respondents, but are desired by quite
many respondents (28%). The age of the respondent does not seem to
influence use of rear-view mirrors. Winter tires and winter cycles are desired by
one fifth of the respondents. However, more than half of the respondents stated
that they do not miss any equipment or that they have no opinion.
The most common sites or maneuvers the elderly avoid are roundabouts, left
turns and crossing streets without a cycle crossing. Also according to the
analysis of Finnish in-depth crash data left-turns were hazardous to the elderly
cyclists. Especially the oldest respondents state that they avoid roundabouts.
Also cycle tracks with moped traffic are avoided by many. The most common
reason that the elderly avoid any site or maneuver is that they feel insecure.
Many choose to walk their bike, when they perceive something dangerous such
as drivers of cars that do not stop or take cyclists into consideration and cars
and mopeds that are driven too fast. However 41% of the respondents do not
avoid any site or maneuver.
According to the elderly, the biggest safety problems are potholes, slipperiness
and bad snow removal; 76, 74 and 70% of the respondents have referred to
these factors as safety problems. However, according to the analysis of the
Finnish in-depth crash data, elderly bicyclists are not over-involved in crashes
where the road surface was damaged. Possibly the explanations are that
elderly ride slower and less in darkness compared to other age groups.
Slipperiness and bad snow removal are problems especially in Southern
Sweden. Major problems are also curb stones and cars going too fast.
One third of the respondents state that signage and route information for
bicyclists is good and another third that it is neither good nor bad. The most
frequent comment about posting of signs is that the quality is varying too much.
It is good at some sites and bad at others. It is sometimes completely missing
and other times damaged.
It is desirable to get information about changes in rules and other news
important to cyclists according to almost a third. Especially respondents older
than 80 years state that information is important. However, Maring and Schagen
[5] conclude that older bicyclists (60+), were deficient regarding knowledge
while showing the most positive attitudes. The subjects over 70 performed
much worse than the rest of the older group concerning knowledge.
What the elderly say would increase their biking is linked to what they say is
important for increased traffic safety. Increased safety would lead to increased
biking among the elderly. Requests dealing with the physical design of roads
are especially a demand for more and better cycle tracks. Communication
256
Diversity and specificity of road user groups
between road users expressed as more and better consideration are also
perceived to increase their feeling of security and thereby increase their biking.
4 Expert questionnaire
An expert questionnaire was distributed during the Velo-city 2007 conference.
All together, 14 experts answered. At the outset the experts were asked to
describe, in their own words, the preconditions for using the bicycle as a means
of transport. The most common preconditions mentioned were: safety and a
feeling of security when cycling, the existence of a network of roads for cycling
including appropriate bike parking facilities and positive attitudes from users and
non-users. This is much in accordance with the opinions expressed by the
senior cyclists. Some experts stressed the importance of an urban policy for
cycle mobility. Reasonable physical and mental abilities of the cyclists were also
considered as important preconditions.
According to the experts, the most important needs concerning infrastructure for
senior citizens are comfortable, wide bike paths or cycle streets away from main
streets, with good directional signage. High curb stones and steep gradients
should be avoided. An electric motor could be useful up-hills. Many experts
mentioned the importance of detectors well in advance of signalized
intersections to give cyclists the possibility to get a green light without having to
slow down or dismount their bicycles.
Low motor vehicle speeds achieved by Intelligent Speed Adaptation (ISA) or by
other means was by many considered as a prerequisite for safety. Other
suggestions to increase safety include warning signals or warning lights to warn
cyclists of approaching motor vehicles or vice versa at intersections. Such
warning devices could also be useful when a motor vehicle is approaching a
bike from behind (or a bike is approaching a pedestrian, but then the sound has
to be “gentle” so that pedestrians are not scared). ITS can be used to get better
guidance for and visibility of bicyclists at night time, for example through ledlights in the pavements or by increasing the intensity of street lighting at times
when cycle traffic is present.
With respect to suggestions to improve the design and equipment of the bike
itself, an upright seating position and a low bike frame making it easy to climb
on and off the bike was stressed. Some equipment facilitating turning left would
be useful as many senior citizens have a stiff neck and bad balance. A rearview mirror could help, as stated by senior cyclists, but improvements are also
possible by designing the infrastructure, so that it becomes unnecessary to
merge with motor vehicles when turning left. As mentioned above, cycle tracks
are an efficient means to increase safety for elderly bicyclists, as they reduce
accidents with left-turning bicyclists [9].
Almost all experts suggested a digital map for on-line route guidance when
cycling and also for trip planning before the trip starts. On-line devices like
Personal Digital Assistants (PDAs) could also be used, for example, to get local
weather information or to find time tables for public transport and especially to
see whether it is allowed to bring the bike on the tram or bus. A special design
of the devices making it easy for elderly to use them was considered crucial.
257
Human Centred Design for Intelligent Transport Systems
The following automatic types of equipment for bikes were considered important
to test and further develop: automatic locking and opening e.g. at a distance by
using the key as for cars, automatic gears, automatic turning on and off of
bicycle lamps (with power supply from a reliable dynamo) and automatic
elevating of the saddle after mounting.
5 An expert workshop
The last tool (the expert workshop) included two group discussions structured
according to two philosophically different models: The Diamond model and The
Multiple comfort model.
5.1
The Diamond model
The Diamond model proposed by Risser [10] includes five areas from which
behaviour-steering effects originate and it mirrors also the fact that effects, or
areas, are interrelated, see Figure 1. Risser and Ausserer [11] argue that traffic
safety experts cannot take decisions that will be accepted by relevant groups,
and they certainly will not get their co-operation, without communicating with
them in an appropriate way.
Individual
(psychology)
Communication,
(social psychology)
Society/ Structures
(sociology)
Mode, “vehicle”
(technology, psychology,
sociology)
Infrastructure
(technology, psychology,
sociology)
Fig.1. The Diamond (Risser, 2000)
The following individual measures were proposed by the group members to
develop safe and joyful cycling for senior citizens: Training, Bike pooling and
Information & instruction.
Concerning the bicycle e.g.: Easily handable lock, Telematics (GPS), Reflectors
and other means to improve visibility, and Assistance for all kinds of communication (e.g. rear mirrors, side blinkers, on-line route guidance).
Concerning infrastructure e.g.: Infrastructure to increase bicyclists' awareness
of pedestrians and car drivers' awareness of bicyclists, incl. infrastructure-based
258
Diversity and specificity of road user groups
telematics, Awareness raising infrastructure design, including blinking lights,
red-coloured lanes, intelligent traffic lights, Sign-posting – big letters and
consistent, Route guidance by signs, telematics (GPS), Give space to bicyclists
and pedestrians, Places to rest, and Transport on public means. .
Concerning society and structure e.g.: Include knowledge about cyclists' needs
and characteristics in driving school curricula, Information of the public about
rules, health issues, Creating a positive image in the media, Change rules and
regulations, and Focus law enforcement on problems of unprotected road
users.
And concerning communication between road users e.g.: Infrastructure
measures (slow down cars, give place for communication), Laws and
regulations enhancing and securing communication, Equipment (rear mirror,
Chinese bell), Training and workshops,
5.2
The Multiple comfort model
There are a lot of models to explain driver behavior. Wilde [12] argues that on
an aggregated level road users tend to target a certain level of risk (risk
homeostasis). This target level of risk can be modified by rewarding safe road
user behavior. Summala [13] argues that the following five issues are the most
important ones to explain driver behavior on a strategic, tactical and operational
(individual) level: safety margins (to survive), good or expected progress of
trips, rule following (according to the law and social rules), vehicle/road system
(bicycle and infrastructure) and pleasure of driving and pleasure of cycling.
The model is slightly modified to also fit to explain cyclists’ behavior. As already
mentioned the second group discussion was structured according to this model.
Safety margins imply a concept of available time, which is one basis for the
behavioral adaptation phenomenon. One example of this phenomenon is that
the safety effect of raising bicycle crossings implying reduced vehicle speeds,
were more or less canceled out by increased bicycle speeds [14]. According to
the group members it is important to make cyclists and cars visible, for example
through warning lights in the pavement, which also was suggested by the
experts in the questionnaire. Otherwise, especially in darkness, safety margins
tend to be insufficient.
Good or expected progress of trips applies also for cyclists. Cyclists like to
maintain their speed and are often hesitating when it comes to braking. As
mentioned above many experts mentioned the importance of detectors well in
advance of signalized intersections to give cyclists the possibility to get a green
light without having to slow down or dismount their bicycles. Gradients,
especially downhill, are hazardous especially for cycling children as they were
reluctant to brake [15]. Group members suggested rumble strips for cyclists to
reduce their speeds at hazardous locations. Cycle infrastructure has to be nonrestrictive to be attractive, see [16].
The analysis of Finnish in-depth crash data revealed that 80% of the cyclists
had not obeyed some rule. Though this figure is certainly biased due to the fact
that the conclusions are often based on the surviving car driver’s statements,
rule following is obviously critical also for cyclists. Group members suggested
259
Human Centred Design for Intelligent Transport Systems
discussions within authorities about rules, for example concerning who should
have the right away: driver or cyclists. They also stressed that clear laws for
cyclists are crucial to facilitate rule following.
According to Summala [13] the vehicle/road system for cars usually implies
smooth car/road performance. This is often not the case for the cycle/road
system. A cycle design for elderly cyclists based on new technology is lacking.
Adequate bicycle infrastructure is often missing in Europe, except in the
Netherlands and Denmark, and if it exists, it often does not comply with the best
practise [16]. Cycle tracks in urban areas should be designed one-directional
[17, 18, 19]. The quality of sign posting for cyclists is varying as stated above.
This is not the case for motor vehicles. However, the most stated safetyincreasing measure according to the senior cyclists is construction of more
cycle tracks. More than one third of the respondents want this. Also according
to research, cycle tracks are an efficient mean to increase safety for elderly
bicyclists, as they reduce accidents with left-turning bicyclists. Jensen [9]
concludes that elderly bicyclists (65+) had a significant reduction in injuries, of
about 55%, when one-directional cycle tracks (with truncated cycle tracks or
raised cycle crossings) were constructed in Copenhagen, though risk increased
by 12% if all age groups were included in the analysis. Both European and
American experiences show that bicycle facilities promote biking [20, 21, 22].
Pleasure of driving will be pleasure of cycling, which obviously is an important
topic for senior cyclists as 84% of the respondents stated that joyfulness is a
reason for them to cycle. Measures should keep or increase the pleasure of
cycling.
6 Conclusions and discussion
All tools tested here seem to work well together for developing ideas for
countermeasures that ensure safe and joyful cycling for senior citizens. With
one exception, all aspects mentioned in the expert questionnaire were taken up
in group discussions in the expert workshop.
Probably, Intelligent Speed Adaptation on cars is the most efficient measure to
provide safe cycling, but other ITS measures are also needed to provide safe
and joyful cycling for senior citizens and raise the profile of cycling as such. ITS
measures could be linked to, or built into, existing equipment such as navigation
systems, cycle computers, and traffic signal control boxes. ITS measures could
also increase the comfort for elderly cyclists, e.g. automatic locking and opening
of bicycles at a distance by using the key as for cars with remote-controlled
locks.
7 Acknowledgement
The research presented in the current paper was done as a case study within
the HUMANIST post-doc grant to the author. The case study was done in
cooperation between VTT, Luleå University of Technology and Factum, and it
was sponsored by Skyltfonden of the Swedish Road Administration, SRA and
the HUMANIST post-doc grant. The grant included a post-doc exchange
260
Diversity and specificity of road user groups
fellowship at Factum, Vienna for three months in the spring of 2007 and another
three months in the autumn of 2007. This was a great experience for me and I
would like to express my sincere gratitude to Ralf Risser, Christine Chaloupka
and all other members of the Factum staff for all advice, help and kindness. Part
of the stay in Vienna was sponsored by Åke och Greta Lissheds stiftelse.
A draft of this paper was discussed at the HUMANIST TFC conference on Joint
Cognitive Models for Human Centred Design in Vienna in September 2007 and
also at an expert workshop before the conference at Factum. I would like to
express my sincere gratitude to all participants and especially those
participating in the expert workshop: Ralf Risser, Karin Ausserer, Corinne
Brusque, Christine Chaloupka and Christine Turetschek.
I would also like to thank Peter Rosander and Mikael Lyckman both working at
Luleå University of Technology, who interviewed the elderly bicyclists by phone
in the pilot project to develop and test the questionnaire to the elderly bicyclists.
I would also like to thank Sven Lindoff, secretary of the Cycling Promotion in
Sweden (Cykelfrämjandet) for helping us with finding a stratified sample of
members for the questionnaire and also for executing the mailing. I would also
like to express my sincere appreciation to all sponsors including Anita Ramstedt
at Skyltfonden, part of the Swedish Road Administration (SRA) for financing
part of the research, and all colleagues, bicyclists and the following experts for
giving fruitful information to achieving safe cycling for elderly: Allessandro
Drago, Italy; anon. colleague from Ireland; Thomas Krag, Denmark; Per Gårder,
United States; Karin Ausserer, Austria; Geoff Gardner, the UK; Leif Jönsson,
Sweden; Sören U. Jensen, Denmark, Christer Hydén, Sweden; Michael
Meschik, Austria; Michal Beim, Poland; Antero Naskila, Finland; and Dirk
Boenke and Jürgen Gerlach, Germany; and to Per Gårder also for proofreading
a draft of this paper; and last but not least to my daughter Laura Leden for
coding, analyzing and reporting the questionnaire to the senior cyclists.
8 References
[1]
[2]
[3]
[4]
[5]
Kraft, K.S. (2002). Report on blood pressure control and on the diagnostic
value of bicycle exercise in elderly people. Medizinische Poliklinik Bonn,
Germany.
Manetta, J., Brun, J.-F.,Prefaut, C. and Mercier, J. (2005). Substrate
oxidation during exercise at moderate and hard intensity in middle-aged
and young athletes vs. sedentary men. Metabolism Clinical and
experimental Vol. 54, pp. 1411-1419.
Oja, Vuori and Paronen (1998). Daily walking and cycling to work: Their
utility as health-enhancing physical activity. UKK Institute for Health
Promotion Research, Tampere, Finland.
Gustafsson, S., Thulin, H. (2003). Gående och cyklister, exponering och
skaderisker i olika trafikantmiljöer för olika åldersgrupper. VTI meddelande
928.2003.
Maring, W. and Van Schagen, I. (1990). Age dependents of attitudes and
knowledge in cyclists. Accident Analysis and prevention, Volume 22 issue
2, pp. 127-136.
261
Human Centred Design for Intelligent Transport Systems
[6]
[7]
[8]
[9]
[10]
[11]
[12]
[13]
[14]
[15]
[16]
[17]
[18]
[19]
[20]
[21]
Leden, L. (2007). Safe and Joyful Cycling for Senior Citizens. Espoo, VTT
Working paper.
Goldenbeld, C. (1992). Ongevallen van oudere fietsers in 1991 [Accidents
of older cyclists in 1991]. R-92-71. SWOV Institute for Road Safety
Research.
Leden, L. and Risser, R. (2007). Safe and joyful cycling for senior citizens.
Velo-city 2007.
Jensen, S. U. (2006). Personal information.
Risser R. 2000, Measuring influences of speed reduction on subjective
safety, Proceedings of the ICTCT Workshop on Traffic Calming, New
Delhi, Vienna.
Risser, R. and Ausserer, K. (2007). From road user models to a
comprehensive marketing approach. Vienna, Factum. Draft.
Wilde, G.J.S. (1994). Target Risk. Dealing with the danger of death,
disease
and damage in everyday decisions. http://psyc.queensu.ca/target/
Summala, H. (2005). Towards understanding driving behavior and safety
efforts. Proceedings of the international workshop on modeling driver
behavior in automotive environments.
Leden. L., Gårder, P. and Pulkkinen, U., 2000, An Expert Judgment Model
Applied to Estimating the Safety Effect of a Bicycle Facility. Accident
Analysis and prevention, Volume 32 issue 4, pp. 589-599.
Leden, L., 1989. The Safety of Cycling Children. Effect of the street
environment. Espoo, VTT. Publications 55. PhD thesis.
Leden, L., 1999. Towards Safe Non-restrictive Cycling. Traffic safety on
two continents. 10th International Conference Sept. 20 – 22, 1999 in
Malmö, Sweden. VTI conference 13 A part 5, Vulnerable Road Users,
1999.
Linderholm, L., 1992. Traffic safety evaluation of engineering measures Development of a method and its application to how physical layouts
influence bicyclists at signalized intersections. Department of Traffic
Planning and Engineering. LTH. Lund.
Summala, H., Pasanen, E., Räsänen, M. and Sievinen, J., 1996. Bicycle
accidents and drivers’ visual search at left and right turns. Elsevier,
Accident Analysis and prevention, Vol. 28. No. 2, pp. 147-153.
Räsänen M., Summala H. and Pasanen E. (1998) The Safety Effect of
Sight Obstacles. and Road-Marking at Bicycle Crossings. Traffic
Engineering & Control. Vol. 39 pp. 98-105.
Nelson, A.,C. & Allen, D., 1997. If you build them commuters will use them.
Association between bicycle facilities and bicycle commuting, Washington,
Transportation Research Record No. 1578. Pedestrian and Bicycle
Research 1997.
Nettelblad, P., 1987. Olycksrisken ökar med dubbelriktade cykelbanor. En
undersökning av 11 cykelbanor i Malmö. (Accident risk increase with two
directional paths. An investigation of 11 cycle paths in Malmö.) Malmö,
Malmö gatukontor.
262
Diversity and specificity of road user groups
[22] Gårder, P., Leden, L. & Pulkkinen, U., 1998. Measuring the Safety Effect of
Raised Bicycle Crossings Using a New Research Methodology.
Washington, TRB, (77nd Annual Meeting Paper No 98-1360), Transport
Research Record No. 1636.
263
Diversity and specificity of road user groups
PRELIMINARY RESULTS OF A STUDY ABOUT THE
INFLUENCE OF AN ACTIVE ACCELERATOR
PEDAL ON NOVICE AND YOUNG DRIVERS IN
AUSTRIA AND THE CZECH REPUBLIC
Christine Turetschek (FACTUM OHG, Austria)
christine.turetschek@factum.at
ABSTRACT: In this presentation preliminary results of a PhD Thesis carried out
in the framework of the NoE HUMANIST will be presented. The thesis itself
concentrates on novice and young drivers, their driving behaviour and the
possibility to influence them towards a safer driving behaviour. This paper will
deal with some descriptive statistics about the first answered questionnaires
and the first results about attitudes towards advantages and disadvantages of
ISA directly after the use of the system and some months later.
1 Objectives
Novice and young drivers of the age from 15 to 19 are in many countries the
group with the highest accident risk in road traffic. In Austria for this age group
the risk to get involved in an accident is three times as high as for the general
population [15]. Adolescent and young adults are those who are injured and
killed mostly on Austrian roads [16]. Additionally it can be stated, that badly
adapted speed is the main reason for their accidents [14].
These facts have been the main reasons to focus on this specific driver group,
novice and young drivers aged from 18 to 30, and on Intelligent Speed
Adaptation, in combination (ISA, see page 2). It is of general interest to learn
whether there are methods to make young drivers safer drivers, for instance
with the help of telematic devices and/or specific training measures.
2 Methods
The study was based on three important types of methods; driver behaviour
observation, psychological group training and questionnaires. The first step was
a behaviour observation of novice and young drivers between the age of 18 to
30, using the method of the “Wiener Fahrprobe”, while driving a driving school
car along standardise routes in Vienna (Austrian novice and young drivers) and
in Brno (Czech novice and young drivers). The observed behaviour at this point
can be seen as a kind of baseline reflecting “usual” driving behaviour of the
participants.
The “Wiener Fahrprobe” is a standardised observation method that gives a
structured impression of an observed individuals driving behaviour. The
subjects were observed by one person sitting in the back of the car behind the
front passenger seat. The route included sections of different road types, such
265
Human Centred Design for Intelligent Transport Systems
as rural roads, motorways, inner city roads, etc., and was divided into sections.
For each section the observer had to fill in a sheet concerning standardised,
more descriptive variables like lane keeping, behaviour at traffic lights, etc., and
non-standardised variables, mostly referring to types of behaviour which could
not be predicted a priory for a certain time and space, like communication with
other road users.
Additionally, drivers had to fill in four questionnaires. One of them, called
“questionnaire for the ISA-study” was developed especially for this study and is
based on recent literature [4, 5, 6, 8, 9 & 10], on informal expert interviews, but
also on two in-depth interviews with persons belonging to the target-group.
Participants had to report about their attitude concerning traffic in general, traffic
safety, speeds and ISA. Furthermore the Manchester Driving Behaviour
Questionnaire [6] which contains questions about the driving style, risks and
mistakes was applied. Another questionnaire was about the type of drivers the
subjects themselves though they belonged to. It was developed by the
psychonomics AG for the Axa Gruppe and can be found at
http://www.autofahrertypen.de/ [17]. Participants had to answer questions about
emotions which are important, their attitudes towards the vehicle, which motives
are related to driving and so on. The last questionnaire was administered
directly in combination with the psychological group training and thus can be
seen as an evaluation instrument.
Some months later all participants drove an ISA (Intelligent Speed Adaptation)
equipped car along the same standardised test routes. ISA devices have been
studied for more than 15 years mostly in Sweden and safety effects are
estimated [2, 3 &11]. In those studies it was often pointed out that ISA has a
positive influence on the communication of the drivers with other road users.
One explanation for this phenomenon is that because of the reduced speed
drivers do have more resources to concentrate on the environment, including
the social environment. The system used in the study reported here was an
Active Accelerator Pedal that gives feedback to the driver by means of pedal
resistance as soon as the legal speed limit is reached [12 &13]. During the ride
with the ISA equipped car drivers were observed with the help of the “Wiener
Fahrprobe”. Again they had to fill in questionnaires; the “questionnaire for ISAstudy” and the one about the type of driver on believes one belongs to.
As a next step, a psychological group training was developed where half of the
drivers should participate. The goal of the training was to shape participants'
opinion concerning traffic safety in general, speed and speeding but also
concerning ISA and the advantages/problems connected to the system. Two
sessions were held within two weeks, each of them lasting two hours. The first
session focused on motives related to driving and emotions associated to the
car. As a consequence of the discussion young drivers should become aware of
discrepancies between different motives, like e.g. the need for safety on the one
hand and the wish to have fun on the other hand. The second session focused
on speed aspects like speed limits, reasons for speeding, problems of
speeding, on the advantages and problems of the ISA system, etc..
Finally the driving behaviour of all participants in Vienna and Brno was again
observed during a last test ride on the standardised route, so that the practical
part of the PhD was finished in August/September 2007.
266
Diversity and specificity of road user groups
2.1
Hypotheses
Different Hypotheses have been developed in the frame of this study, mainly
based on the assumptions that have been made and discussed before,
completed by the preparation work carried out here. The main ones were the
following:
Hypothesis 1: The use of an ISA system will improve the communication
between the driver and other road users.
Hypothesis 2: Only short-term use of ISA will not be enough to achieve a long
term effect on the drivers' attitudes and behaviour.
Hypothesis 3: Special training programs as envisaged in connection with the
PhD thesis presented here can change negative attitudes towards ISA and
towards speed behaviour in general.
Hypothesis 4: Such training measures have the potential to improve drivers'
attitudes and behaviour for a longer period of time.
2.2
Limitations in this paper
There have to be mentioned two limitations of this paper. One limitation is
related to the study design. This PhD thesis started as a very ambitious project,
with many ideas about how to reach these very specific subjects. In reality it
turned out to be more difficult to attract novice drivers to take part in the study.
The real number of participants related to answered questionnaires can be seen
in table 1. In Austria and the Czech Republic different methods turned out to be
successful. In Austria people have been recruited trough a webpage where
students search for jobs. In the Czech Republic thanks to our partner CDV a
driving instructor helped us to search for participants, which was not possible in
Austria. That is the reason, why the Austrian subjects are in average older than
the Czech ones but also why subjects in Austria have already quite a high
driving experience which can be seen in the next chapter. It also has to be
noticed that the test and the control group do not have the same size because
of practical reasons. It was quite difficult to motivate the young drivers to
participate in the group discussion. All results presented here but also those
which will be provided in the future will have to be evaluated with respect to
these constraints.
267
Human Centred Design for Intelligent Transport Systems
Table1. Study design, pictured by an overview of number of answered
questionnaires
Steps of the study
Step 1
Step 2
Step 3
Test group
Number of people in
country
AUT
CZ
Drive in a driving school car
47
27
Fill in "ISA" questionnaire
47
27
Drive an ISA equipped car on
an extra route to get used to
the system
40
23
Drive an ISA equipped car on
the original route
40
23
Fill in "ISA" questionnaire
40
23
Fill in MDBQ
40
23
Fill in "Type" questionnaire
40
23
Participate in the pschological
group training
15
8
Fill in "ISA" questionnaire
15
8
Fill in MDBQ
15
8
Fill in "Type" questionnaire
15
8
Fill in "Training" evaluation
15
8
15/20
8/15
15/20
8/15
15/20
8/15
15/20
8/15
Drive in a driving school car
Step 4
Fill in "ISA" questionnaire
Test group/
Control group Fill in MDBQ
Fill in "Type" questionnaire
Sum
Drop Out
74
0
63
11
23
58
16
The other limitation concerns the results presented here. This paper will not yet
deal with all of the central hypotheses of the research, but will concentrate on
general attitudinal aspects of the subjects whom we have communicated with,
concerning their view on what are safe and unsafe drivers, and how an ISA
system, according to their point of view, could interfere with the traffic system.
The latter aspect was analysed with the help of open-ended questions. The
spontaneous answers are summarised in the result chapter. Additionally a first
analyse of the answers given concerning ISA will be presented in this paper.
2.3
Participants sample
74 novice and young drivers could be motivated to participate but only 58
finished the study. The others dropped out because of different reasons and at
different steps of the study. Out of those 58 just 57 filled in all questionnaires.
34 participants come from Austria and 23 from the Czech Republic. All of them
are aged between 18 and 30 years. The mean age in Czech Republic was 19,
in Austria 23. The distribution of sex was different in both countries. In the
Czech Republic only few women took part whereas in Austria about half of the
drivers were female.
268
Diversity and specificity of road user groups
Five participants did not have a valid driving licence at the moment of the first
ride. 52 had a driving licence allowing them to drive passenger cars and some
of them also had a driving licence for other vehicles, like motorbikes or lorries.
The driving experience with vehicles of category B can be seen in figure 1.
Amount of particpiants with driving licence B
20
18
16
14
12
10
8
6
4
2
0
0-5.000 km
5.000-10.000 10.000-20.00 0 20.000-50.000
km
km
km
50.000100.000 km
mehr als
100.000 km
Fig.1. Driving experience
The driving experience was in average about 42 months although it has to be
said, that the driving experience of the two groups was quite different. In Austria
participants had acquired their licence in average 61 months ago, in the Czech
Republic this was only about 12 months ago.
3 Results
Concerning their self-perception most novice and young drivers experience
themselves as safe drivers but not all of them feel safe in traffic. In the Czech
Republic participants feel less safe than in Austria. Good drivers according to
participants' view are those, who behave according to the law, who drive
anticipatorily and safely, while drivers who endanger others, drive aggressively,
do not behave according to the law, drive too fast and reckless. According to
the interviewed subjects, speed limits are necessary and are considered as
"quite ok" in both countries.
Adolescents in this study have not been aware of ISA that much but all of them
have been able to think about advantages and problems of such a system. The
most frequent statements (given spontaneously) that reflect advantages of the
system were:
•
the system makes the driver aware of speed limits
•
gives possibility to control one’s behaviour
•
can increase traffic safety.
The most frequent disadvantages assumed by novice and young drivers were:
•
ISA could be a handicap if someone wanted to overtake
269
Human Centred Design for Intelligent Transport Systems
•
the system would take too much control so that the driver could not behave
as he/she likes
•
there could be some delegation of responsibility to the system, like
speeding when the system was out of function, or speeding in areas where
the system would not work
•
the system might produce possible errors like setting an erroneous limit,
etc.
However, some of the participants believe that ISA does not have any
disadvantages at all. All in all, the young drivers that we communicated with are
rather neutral concerning the question whether they would use such a system
or not. At this point of time it is legitimate to say that their attitude to ISA is not
clearly positive, but it is definitely not negative either.
As a next step, the attitudes of novice and young drivers regarding ISA shortly
after the use of the system (step 2) compared to their attitude about one year
later (step 4) distinguished between the group of subjects with (test group) and
without (control group) participating at the group discussion will be reported
here as well. The test group consists of 23 subjects whereas in the control
group 35 subjects participated. This difference resulted because of practical
reasons which already have been mentioned above. It can be seen in figure 2
and 3 that all novice and young drivers assess the advantages of ISA quite
positively. On a 5 point scale there are only one to two advantages that are
assessed with a value higher than 3 in average (“No control look on
speedometer necessary.” and “No control look on traffic signs necessary”).
Within the control group, people who did not take part in the group discussion,
there are no significant differences between the answers given directly after
using the ISA system and those given about one year later after the last test
ride. Those participants who underwent the group discussions (test group)
assess the advantage “Making aware of speed limits” significantly (t=-2,806;
df=21; p=0,011) better directly after using the ISA system (step 2). This result
goes against hypothesis 3 at a first step. Further work will have to concentrate
on the reasons of that finding.
270
Diversity and specificity of road user groups
totally correct
totally incorrect
Making aware of speed limits
Clear information about permissible maximum speed
Feedback/reminder on speed limits
Warning about transgression of the allowed speed limits
Prevents unintentional speeding
Speed regulation
Support for the driver
More concentration on the traffic possible
Increases the traffic safety
Control general
No control look on speedometer necessary
No control look on traffic signs necessary
1
2
3
Point in time 1
4
5
Point in time 2
Fig.2. Advantages of ISA without group discussion
totally correct
totally incorrect
Making aware of speed limits
t=-2,806; df=21; p=0,011
Feedback/reminder on speed limits
Clear information about permissible maximum speed
Prevents unintentional speeding
Warning about transgression of the allowed speed limits
Support for the driver
Speed regulation
More concentration on the traffic possible
No control look on speedometer necessary
Increases the traffic safety
Control general
No control look on traffic signs necessary
1
2
3
Point in time 1
4
5
Point in time 2
Fig.3. Advantages of ISA with group discussion
It is about the same for the disadvantages, shown in figure 4 and 5. Most of the
participants agree on the mentioned disadvantages of the ISA system. Within
the answers regarding to disadvantages of ISA there is a quality difference. For
271
Human Centred Design for Intelligent Transport Systems
instance to realise that ISA might make people rely on the system too much
indicates awareness of possible dangerous results. In contrast to this, if ISA is
seen as an annoying system it shows that one does not want to renounce fun in
order to get safety. In the control group there is one significant difference
(t=2,272; df=31; p=0,030) in their rating concerning the reliability of the system.
Subjects answered in the last questionnaire (step 4) that ISA users might rely
too much on the system, which they thought directly after the use of ISA (step
2) as well. On the other hand there are five significant differences in the
attitudes between the two points in time within the test group again. Just two of
them “One relies too much on technology. “ and “Uncertainty, if one is
unaccustomed to this.” are in the expected direction. The other three
statements suggest that the attitudes of the test group towards ISA had
deteriorated despite the group discussion in comparison to the control group.
totally correct
totally incorrect
Those which want to drive too fast doing this in spite of such systems
Missing up-to-dateness of the system
Can sometimes be in the way
Occurrence of a defect
One relies too much on technology
Uncertainty, if one is unaccustomed to this
Restriction of the driver
Annoying/frustrating/bothersome
Too much control
Paternalism
Purchase too expensive
Less concentration on the traffic situation/distraction
1
2
3
Point in time 1
Fig.4. Disadvantages of ISA without group discussion
272
4
Point in time 2
5
Diversity and specificity of road user groups
totally correct
totally incorrect
Those which want to drive too fast doing this in spite of such systems
Missing up-to-dateness of the system
Can sometimes be in the way
t=2,982; df=21; p=0,007
One relies too much on technology
Purchase too expensive
t=2,238; df=21; p=0,036
Occurrence of a defect
Restriction of the driver
Too much control
t=2,421; df=21; p=0,025
t=2,658; df=21; p=0,015
Annoying/frustrating/bothersome
Paternalism
Uncertainty, if one is unaccustomed to this
t=2,215; df=21; p=0,038
Less concentration on the traffic situation/distraction
1
2
Point in time 1
3
4
5
Point in time 2
Fig.5. Disadvantages of ISA with group discussion
4 Conclusions
It was found that novice and young drivers in our sample do feel quite safe in
traffic and also have many different ideas about attributes of a safe driver and a
driver who endangers others. They also have a good imagination about what
ISA could accomplish and where the limitations of such a system could lie.
There can be given just few and very restricted answers to hypothesis 3. So far
it must be stated that the type of group discussion, realised within this study, did
not work in the right way. A reason for this findings might be a group effect of
the test group, or it could be that a sensitisation of the test group in regard to
problems ISA might have taken place. But this new hypothesis has to be
discussed in further steps of analysis. For instance, a comparison of both
groups concerning each point in time, broken down into step 2 and 4, will help
to find out whether there are significant differences between the test and the
control group.
So far it can be stated that a very restricted group discussion in terms of time,
as implemented in the frame of the thesis, ad-hoc does not have the potential to
change the attitudes towards more traffic safety of novice and young drivers.
Maybe this group training at least can cause a sensitisation. Nevertheless for
future work a different, “more sophisticated” type of group work has to be
considered. Especially, it will have to be more extended in time, as also asked
for in connection with legally applied group procedures in, e.g., Austria and
Germany [1].
273
Human Centred Design for Intelligent Transport Systems
Concerning ISA, in contrast to earlier results, attitude did not change in a
positive way after the use of ISA, and those who took part in the group sessions
even became more negative, at least in some argument. This might be an effect
of age, where one may assume that the attitude towards ISA generally is not
that positive and that discussion in the groups even could produce reactance.
To provide comparisons of different nationalities was not possible, due to the
differences of the two groups, but this actually never was planned, although it
would be interesting. But in any case, the study can provide insight into the
structure of attitudes of young drivers and how they can be influenced. But this
will only be possible after final provision and analysis of data.
5 References
[1]
Bartl, G., Assailly, J.P., Chatenet, F., Hatakka, M., Keskinen, E. & WillmesLenz, G; EU-Project „Andrea“ – Analysis of Driver Rehabilitation
Programmes. Kuratorium für Verkehrssicherheit. Wien. (2002).
[2] Carsten, O., and Comte, S.; UK Work on Automatic Speed Control.
Proceedings of the ICTCT 97 conference. 5-7 November 1997, Lund,
Sweden. (1997).
[3] Carsten, O., and Fowkes, M.; External Vehicle Speed Control. Executive
Summary of Project Results. University of Leeds, UK. (2000).
[4] Engström, I., Gregerson, N.P., Hernetkoski, K., Keskinen, E. & Nyberg, A.;
Young novice and youngdrivers, driver eductaion and training – Literature
review; VTI rapport 491A. (2003).
[5] Gregerson, N.P.; Young drivers´overestimation of their own skill an
experiment on the relation between training strategy and skill; Accident
Analysis and Prevention, 28(2), 243-250. (1996).
[6] Lamszus, H.; Die Problematik junger Fahranfänger und Möglichkeiten zur
Verringerung ihres hohen Unfallrisikos; Zeitschrift für Verkehrssicherheit,
Heft 3, Seite 121; TÜV-Verlag GmbH, Köln. (2002).
[7] Manchester Driving Behaviour Questionnaire. (2002).
[8] Mienert,
M.;
Merkmale
potenzieller
Risikofahrer
vor
dem
Führerscheinerwerb; Zeitschrift für Verkehrssicherheit, Heft 4, Seite 145;
TÜV-Verlag GmbH, Köln. (2002).
[9] Mienert, M.; Entwicklungsaufgaben Automobilität – Teil1; Psychische
Funktionen des Pkw-Führerscheins für Jugendliche im Übergang ins
Erwachsenenalter, Heft 1, Seite 26, TÜV-Verlag GmbH, Köln. (2003).
[10] Mourant, R.R. & Rockwell, T.H.; Strategies of visual search by novice and
youngand experienced drivers; Human factors, 14, 325-335. (1972).
[11] Várhelyi, A.; Dynamic speed adaptation based on information technology a theoretical background. PhD Thesis, Lund University, Sweden. (1996).
[12] Várhelyi, A., Hydén, C., Hjälmdahl, M., Almqvist, S., Risser, R., and
Draskóczy, M.; Effekterna av aktiv gaspedal i tätort. Sammanfattande
rapport. LundaISA. (The effects of active accelerator pedal in built-up
areas. Summary report) (In Swedish) Bulletin 210. Lund University,
Sweden. (2002).
274
Diversity and specificity of road user groups
[13] Várhelyi, A., Hydén, C., Hjälmdahl, M., Draskóczy, M.,and Risser, R.,; The
effects of active accelerator pedal on driver behaviour and traffic safety
after long time use in urban areas, Department of Technology and Society,
Lund University, Sweden. (2002).
Internet
[14] Federal Statistical Office (Statistisches Bundesamt); Unfallgeschehen im
Straßenverkehr 2004; Statistisches Bundesamt; Germany, Wiesbaden.
(2005).
http://wwwec.destatis.de/csp/shop/sfg/bpm.html.cms.cBroker.cls?CSPCHD=0000000
10003liLeA34kJo3106456009&cmspath=struktur,Warenkorb.csp
[15] Kuratorium für Verkehrssicherheit (Hg.); Verkehrsunfallstatistik 2003 –
Jugendliche; Kuratorium für Verkehrssicherheit, Austria, Vienna. (2004)
http://www.kfv.at/ps/tools/download.php?file=/live/kuss/kuss_kfv/psfile/pdfl/
63/Jugendl0341f90b517fba4.pdf&name=Jugendl03.pdf&id=247&nodeid=5
19
[16] Kuratorium für Verkehrssicherheit (Hg.); Verkehrsunfallstatistik 2006;
Kuratorium
für
Verkehrssicherheit,
Austria,
Vienna.
(2007)
http://www.kfv.at/fileadmin/Publikationen/Verkehrsunfallstatistiken/2006/VU
S2006.pdf
[17] http://www.autofahrertypen.de/
275
Diversity and specificity of road user groups
COGNITIVE ACTIVITY MODELLING: A CASE
STUDY OF LANE CHANGE SCHEMAS AND
SENSATION SEEKING
Tony Wynn (ESRI, United Kingdom),
John H. Richardson (ESRI, United Kingdom),
Olivier Georgeon (INRETS, France), Thierry Bellet (INRETS, France),
Matthias J. Henning (Chemnitz University of Technology, Germany),
Josef F. Krems (Chemnitz University of Technology, Germany)
T.wynn@lboro.ac.uk
ABSTRACT: We propose a method to investigate differences of driving at a
tactical level, between different categories of drivers. We analysed the
naturalistic driving performance of 19 French drivers using an instrumented
vehicle. We assessed their sensation seeking scores with the Zuckerman
questionnaire. We observed a significant correlation between their sensation
seeking score and their mean speed on motorway. We set up a method to
model their tactical behaviour and investigate possible correlation between their
sensation seeking score and their tendency to perform certain types of
behaviour. We applied it to the study of lane changes on motorways. We could
model two categories of lane changes but we show that they were not
correlated with the sensation seeking score. Despite this negative first result,
we are proposing an innovative approach for this kind of study.
1 Introduction
Sensation seeking is a personality trait defined by the seeking of varied, novel,
complex, and intense sensations and experiences and the willingness to take
physical, social, legal and financial risks for the sake of such experiences [1].
The driving environment is an everyday situation which allows sensation
seekers to exhibit typical sensation seeking behaviours. The automobile can
provide a variety of intense experiences including: speeding, racing other cars,
reckless overtaking [2]. High sensation seekers are more likely to have
accidents and receive traffic citations [3, 4] and are significantly more likely to
report aggressive driving habits [4].
The habits and preferences of high sensation seekers (HSS) are aimed at
maintaining an optimal high level of arousal. Driving is a significant means of
achieving this. It is reasonable to assume that accident and speeding ticket
prone young drivers are more likely to be high sensation seekers, scoring high
on scales that measure attitudes towards speeding and harbouring self-serving
attitudes and erroneous beliefs congruent with their higher preferred levels of
arousal [5, 6, and 7]. [8] indicated that high sensation seekers were significantly
more likely than low sensation seekers to speed, not wear seat belts, drink
frequently, drive after drinking, perceive a low risk of detection for impaired
driving and perceive that they could drink more beer after being impaired. [9]
illustrates that those individuals who score highly on scales which measure
277
Human Centred Design for Intelligent Transport Systems
sensation seeking are also more likely to commit traffic violations, are more
impulsive and are less likely to use their safety belts.
It is apparent that there are substantial differences in the driving behaviour
observed in high and low sensation seekers. Questions remain however
regarding the differences in the cognitive processes that govern the behaviour
of sensation seekers and how these differences can be identified? In an attempt
to answer this question we apply the ABSTRACT cognitive modelling approach
to data collected using the instrumented vehicle. Firstly we aim to establish that
the sample we have selected is representative of sensation seeking behaviour
by correlating behaviours that typify the sensation seeker (average speed,
maximum speed and Time-To-Collision) with participant’s scores on the general
sensation seeking scale. We then used the ABSTRACT methodology to
develop a case study comparison to investigate the driving performance of high
and low sensation seekers in order to establish differences in cognitive
schemas between high and low sensation seekers.
The cognitive modeling of an operator is based on the fundamental notion that
they do not act according to the objective status of the world but according to a
"mental model" or "representation" that they make of it. Cognitive models are
usually constructed by cognitive psychologists or ergonomists. They rely on
concepts such as “schemas”, "Scripts" [10], "Scenarios", "Operative Image"
[11], or “Situation Awareness” [12]. The latter offers a covering notion by
simultaneously addressing the perception, the comprehension and the
anticipation for describing how the elements of the situation are “mentally” taken
into account by the operators.
The question of knowing which data must be selected to describe “the
meaningful (to the subject) aspects of its activity" is a crucial question when one
builds cognitive models. Answers to this question can come from both an
analysis of the activity and from cognitive theories. To address this question we
have developed a method and a tool named Abstract which is described bellow.
This method is based on the collection of behavioral data from driving with an
instrumented car. And the transformation of this data to a higher level
description of the activity focusing on what is held as being meaningful to the
driver. We can so highlight significant patterns of behavior that were actually
performed by the subject during the experiment. These patterns can be taken
as mental schemas of the activity. It is reasonable to assess that these mental
schemas will depend on the subject's personality. This is this assumption that
we would like to investigate here.
2 Method
2.1
Participants
19 (9 male, 10 female) participants were selected at random from an
opportunity sample of French drivers. All participants held a French Category B
driving license and had normal to corrected vision. The Category B licence
permits a driver to drive a vehicle with a maximum weight of 3.5 tons, and
seating no more than 9 passengers, including the driver. This includes standard
passenger cars, people carriers and microbuses. The average age of
278
Diversity and specificity of road user groups
participants was 33 years old (range 25 to 55) and they had held a Class B
licence for an average of 13 years (range 2 to 39). They reported an average
annual mileage of 12,364 miles per year (range 2,000 to 50,000). Participants
received a standard payment of €80 for taking part in this study.
2.2
Instrumented Vehicle
The experiment is run using an instrumented vehicle which is a Renault Scenic.
It has a two liter, 16 valve engine. It is fitted with four video cameras that record
front view, rear view, side mirrors and the driver and two roof top mounted
cameras for obstacle detection. There are a further two cameras linked to an
‘Eyetracker’ system that are responsible for detecting the ‘zone of interest’
towards which the driver has directed their gaze.
Fig.1. The LESCOT instrumented vehicle.
It is fitted with a Navteq ADASRP global positioning system (GPS) that is used
to compare the recorded data points with the video output. There are four
computers located in the rear storage compartment; 1 Pentium III, 1 GHz, P.C.
dedicated to collecting vehicle parameters (pedals, steering wheel, speed,
acceleration, etc.); 1 Pentium IV 3 GHz dedicated to collecting FaceLab data
(gaze detection, direction, etc.); 1 dual Pentium III 1.6 GHz dedicated to
obstacle detection and 1 Pentium M 1.6 GHz dedicated to GPS information. It
has one point telemeter (LMS) located on the front right of the vehicle and one
scanner telemeter (SICK) in front of the car at the centre that works in
conjunction with two cameras located on the roof of the vehicle for obstacle
detection. The algorithm used in the detection system was developed by
INRETS-LIVIC during the ARCOS project (2004). It is equipped with sensors on
each pedal, the steering wheel, the indicator and headlights [13].
2.3
The course
The course of the experiment is 50 km long and includes urban, suburban and
motorway sections. The urban and suburban sections include numerous
junctions, roundabouts and traffic lights. Participants are required to navigate a
number of different road types ranging form small roads to large motorways.
279
Human Centred Design for Intelligent Transport Systems
This study is primarily concerned with data collected during the motorway
section. During the whole course there is a significant interaction with other road
users.
2.4
Zuckerman’s Sensation Seeking Scale Form V (SSS)
The Sensation Seeking Scale [1] is composed of 40 dichotomous forced choice
questions. The forced choice paradigm requires participants to choose the
option that most reflects their behaviour. A score of one is given if the sensation
seeking option is selected. A score of zero is assigned if the alternative option is
chosen. The total score is generated by calculating the number of sensation
seeking responses. This scoring system is consistent with other applications of
the sensation seeking scale.
Within the general sensation seeking scale there are four contributing factors
[14, 15]. Each of these sub-scales is represented by ten questions on the SSS.
The dimensions of sensation seeking explored by the SSS are; Thrill and
adventure seeking (TAS; Contains questions relating to the participants’
attraction to thrill, dread and particularly a willingness to engage in physically
risky activities or extreme sports like parachuting, mountaineering and skiing);
Experience seeking (ES; Contains questions relating to participants aspirations
to undergo a variety of novel experiences through the mind and senses,
especially arousing music, art, travel, social non-conformity and the association
with fringes of conventional society); Disinhibition (Dis; Contains questions
which relate to participants loss of self control and sensation seeking through
social activities e.g. drinking, parties, sex) and Boredom Susceptibility (BS;
which refers to their intolerance toward monotonous, repetitious or predictable
people and events).
2.5
Method of behavioural data analysis
For the behavioural data analysis, we have used a methodology and a tool
developed at Inrets, called Abstract (Analysis of Behaviour and Situation for
menTal Representation Assessment and Cognitive acTivitiy modeling) [16, 17].
It allows us to translate the data collected from sensors in the instrumented
vehicle into a higher level of description. This higher level of description
highlights patterns of behaviour that can be explained by the fact that the
drivers carries out cognitive schemas. These cognitive schemas are shaping
the know-how and the reasoning performed by the driver.
3 Results and Discussion
The first aim of this study was to analyse the behaviour of participants to ensure
that it was consistent with that described in the sensation seeking literature. A
key indicator of sensation seeking behaviour in the driving task is speed. In this
instance both the average speed and maximum speed were significantly
correlated with total sensation seeking score. The presence of this trend in the
data is evidence that the sample selected is representative of sensation seekers
in general.
280
Diversity and specificity of road user groups
40.00
40.00
30.00
SSST
SSST
30.00
20.00
20.00
R Sq Linear = 0.303
10.00
R Sq Linear = 0.433
10.00
95.00
100.00
105.00
110.00
115.00
120.00
125.00
120.00
Mean_Speed
125.00
130.00
135.00
MAX_Speed
Fig.2. Correlation of SS total score with mean and maximum speed
There is a significant correlation between sensation seeking total score and
mean speed (r (17) = .550, p< .05). This is illustrated in Figure 2 which shows
that as sensation seeking total score increases so does the mean speed
travelled in the instrumented vehicle. A similar trend is observed between total
sensation seeking scores and the maximum speed travelled in the instrumented
vehicle. Figure 2 shows that there is an increase in maximum speed as total
sensation seeking score increases. This is supported by the statistical analysis
of this trend (a significant correlation between sensation seeking total score and
maximum speed (r (17) = .658, p< .01).
A further variable that has been associated with sensation seeking is headway
(measured using time-to-collision; TTC); in this instance headway was
measured by the forward telemeters. A decline in TTC as shown in figure seven
is suggestive of a decrease in the headways maintained by participants scoring
high on sensation seeking and although the results of a statistical analysis
indicate that the relationship between sensation seeking total scores and TTC is
not significant (r (17) = -.341, p> .05). The fact that this is approaching
significance is reassuring given the context of data collection. It is possible that
there is variability in the TTC measure due to inconsistencies in traffic flow.
Differing traffic flow is a key issue in terms of the reliability of results as there
are few controls that can limit the impact of this variable. In this study the time of
day at which the experiment took place was controlled in the hope that peak
traffic flow would be consistent for that particular time, however this does not
take into account the impact of school and national holidays, commuters,
weekend traffic or other factors that may impact the number of road users
during any particular trial.
The weather conditions at the time of the trial can have an impact on the data
collection. Rain or overcast conditions can affect the efficiency of the telemeters
that measure headway. It can also interfere with the clarity of the video
281
Human Centred Design for Intelligent Transport Systems
recordings provided by the in car cameras. Sunshine can also cause problems
with data collection as glare can affect the output of the video recording
equipment and can also impact on the accuracy of the eye-tracking data due to
reflective glare and the ‘squinting’ of participants. The negative elements of
using a naturalistic driving scenario are negated by the rich data that can be
collected. When using an ecologically valid experimental paradigm we can be
assured that the behaviour observed is realistic.
40.00
SSST
30.00
20.00
R Sq Linear = 0.116
10.00
10.00
20.00
30.00
40.00
50.00
TTC
Fig.3. Correlation of sensation seeking total score with time to collision
3.1
Lane change schema
We have focussed our study of tactical behaviour on lane change manoeuvres.
We have chosen this type of manoeuvre because it is ubiquitous in the driving
environment [18]. It combines many critical features of driving including control,
monitoring and decision making. A lane change is defined as a more or less
stable sequence of actions that begins with the motivation to change lanes i.e.
in the presence of a slow leading vehicle, followed by a gathering of information
about the surrounding traffic situation and then the decision whether to change
lanes or not. The analysis with Abstract let us [18] identify the two categories of
lane change schemas that are explained below.
282
Diversity and specificity of road user groups
Fig.4. Lane_Change_Accelerate
Figure 4 shows a lane change with acceleration. In this situation the driver is
impeded by a vehicle slower than his desired speed. He may check his left
mirror several times, then, when he decides to overtake, he accelerates while
checking his mirror; he switches his blinker on, starts steering, and crosses the
line.
Fig.5. Lane_Change_Stable_Speed
Conversely figure 5 illustrates a lane change where speed is stable, in this
situation, the overtaking is done "on the fly" with no speed variation. The blinker
is switched on with a good anticipation before the checking to the left mirror and
the beginning of the manoeuvre.
Can we therefore, identify differences in the steps (cognitive schema) taken by
high and low sensation seekers in a specific driving behaviour (overtaking)? We
aimed to do this by identifying differences between frequencies of lane changes
by participants with divergent sensation seeking scores. Every participant
performs both Lane_Change_Stable_Speed and Lane_Change_Accelerate
lane changes, but we could formulate the assumption that high sensation
seekers would perform more Lane_Change_Accelerate, because they may
have a more dynamic driving style.
However, the relationship between SS-Total and total number of lane changes
was not significant, nor did we find evidence to confirm our hypothesis that
sensation seekers would perform more lane changes with acceleration than
stable lane changes. The lack of a significant relationship between sensation
seeking and lane changes is discouraging. However, these schemas were not
made with the intention to describe typical low or high sensation seeking
patterns of behaviour. We attempted to correlate schema previously defined by
[18]. A future direction for this research may involve the analysis of driving
behaviour in which participants are pre-selected in terms of their sensation
seeking scores. The ABSTRACT methodology could also allow us to model
behaviour from the point of view of Sensation Seeking by focussing on
indicators that we think may highlight differences with regard to sensation
seeking, such as the strength of acceleration or steering.
There are further improvements that could be made to this methodology that
may increase the likelihood that we achieve the hypothesised results, they are
to increase the sample size and use a simulated driving task. With a larger
sample of participants we can increase the power and accuracy of the
parameters that we are measuring [19].It will reduce the variability of measures
which will in turn yield more sensitive hypothesis tests with greater statistical
283
Human Centred Design for Intelligent Transport Systems
power and smaller confidence intervals. Additionally replication of this
experiment in a driving simulator may give us greater control of the
experimental context. As suggested in the discussion of the relationship
between sensation seeking and TTC the differences in road environment (e.g.
traffic flow) may impair our ability to record accurate data. The application of the
ABSTRACT methodology in an advanced driving simulator would allow strict
control of the driving environment while maintaining the ecological validity of the
task. In its current form (instrumented vehicle trial) the Data can be noisy,
meaning that the patterns identified are meaningless. Simulation would allow
us to replicate the exact traffic flow conditions for each participant so that they
are effectively performing the same lane change at the same time which would
allow us to better compare the differences.
Given the lack of evidence to support our hypothesis we may have to conclude
that sensation seeking is not an important influence on driver’s lane change
behaviour. However we believe that we have applied a novel approach to
sensation seeking using an innovative methodology and that with further
development we may be able to disseminate a relationship between sensation
seeking and lane change behaviour.
4 References
[1]
Zuckerman, M. "Dimensions of sensation seeking", 1971, J Consult. Clin.
Psychol. 36: pp. 45-52.
[2] Arnett, J.J., Offer, D. and Fine, M.A.: Reckless driving in adolescence:
‘state’ and ‘trait’ factors. Accid. Anal. & Prev., 1997, 29(1), pp. 57-63.
[3] Furnham, A. and Saipe, J.: Personality correlates of convicted drivers,
Pers. and Ind. Diffs., 1993, 14(2), pp. 329-336.
[4] Heino, A., van der Molen, H.H. and Wilde, G.J.S.:Differences in risk
experience between sensation avoiders and sensation seekers, Pers. and
Ind. Diffs., 1996, 20(1), pp.
[5] Arnett, J.: Heavy metal music and reckless behaviour among adolescents,
J Youth and Adol., 1991, 20(6), pp. 573-592
[6] Clement, R. and Jonah, B.A.: Field dependence, sensation seeking and
driving behaviour. Pers. and Ind. Diffs., 1984, 5 (1), pp. 87-93.
[7] Whissell, R.W. and Bigelow, B.J.:The speeding attitude scale and the role
of sensation seeking in profiling young driver at risk. Risk Analysis, 2003,
23(4), pp. 811- 820.
[8] Jonah, B., Thiessen, R. and Au Yueng, E.: Sensation seeking, risky driving
and behavioural adaption, Acci. Anal. and Prev., 2001, 33(5), pp. 679-684
[9] Wilson, R.J.: The relationship of seat-belt non-use to personality, lifestyle
and driving record. Health Education Research, 1990, 5(2), pp. 175-185
[10] Schank, R. C. and Abelson, R. P.:
Scripts, plans, goals and
understanding. Hillsdale, N.J., 1977.
[11] Ochanine, V. A.: Concept of operative image in engineering and general
psychology. Engineering Psychology. B. F. Lomov, V. F. Rubakhin and V.
F. Venda. Moscow, Science Publisher
284
Diversity and specificity of road user groups
[12] Endsley, M. R.: "Toward a theory of situation awareness in dynamic
systems." Human Factors, 1995, 37(1): pp. 32-64.
[13] Labayrade, R., and Aubert, D.: A single framework for vehicle roll, pitch,
yaw estimation and obstacles detection by stereovision. IEEE Intelligent
Vehicle Symposium 2003, Columbus, USA, 2003, pp. 9-11.
[14] Zuckerman, M., Eysenck, S.B.G., and Eysenck, H.J.: Sensation seeking in
England and America: cross-cultural, age and sex comparisons. J of Cons.
and Clin. Psych., 1978, 46, pp. 139-149.
[15] Rowland, G.L., and Franken, R.E.: The four dimensions of sensation
seeking: a confirmatory factor analysis. Pers. and Ind. Diffs., 1986, 7(2),
pp. 237-240.
[16] Georgeon, O., A. Mille, and Bellet, T.: Analyzing behavioural data for
refining cognitive models of operator. Philosophies and Methodologies for
Knowledge Discovery, Krakow, Poland, IEEE Computer Society Press
Proceedings, 2006.
[17] Georgeon, O., Henning, M.J., Bellet, T. and Mille, A.: Creating cognitive
models from activity analysis: a knowledge engineering approach to car
driver modelling. International Conference on Cognitive Modeling. Ann
Arbor, MI, 2007.
[18] Henning, M. J., Georgeon, O. & Krems, J.: The quality of behavioural and
environmental indicators used to infer the intention to change lanes. 4th
International Driving Symposium on Human Factors in Driver Assessment.
Stevenson, Washington USA, 2007.
[19] Ajgaonkar, S.G.P.: The effect of increasing sample size on the precision of
an estimator. The Amer. Stat., 1967, 21(24), pp. 26-28.
285
Diversity and specificity of road user groups
RISK AWARENESS ANALYSIS: A COMPARISON
BETWEEN CAR DRIVERS AND MOTORCYCLISTS
Aurélie Banet, Thierry Bellet (INRETS, France)
aurelie.banet@inrets.fr
ABSTRACT: The general framework of this research is about risk awareness
through the aspect of cognitive abilities - to perceive and assess the criticality of
the driving situation - among motorcyclists versus car drivers. In order to study
risk awareness, we have developed some tools based on video films. The
experimental task consists in assessing the criticality of dynamic driving
situations, via a Likert scale, and qualifying these situations using Osgood’s
semantic differential. The results show that riders globally consider our sample
of driving situations as less critical than car drivers do. This paper presents the
main differences observed between the two populations.
1 Introduction and objectives
This paper deals with drivers versus riders’ awareness of risk while driving. Risk
is widely studied in scientific literature as it helps explaining a high proportion of
traffic accidents. Risk is then studied from different angles, such as risky driving
behaviour through the study of motivations and attitudes towards risk. First of
all, risk is more or less linked to human errors problems [1]. Driver Behaviour
Questionnaire (DBQ) is one of the tools designed to study this aspect through
verbal methods [2]. Risk taking is defined as either a socially unacceptable
volitional behaviour with a potentially negative outcome in which precautions
are not taken or a socially accepted behaviour in which the danger is
recognized [3]. Risk sensation tries to determine the connection between a risky
driving behaviour and the driver’s personality factors. Sensation Seeking Scale
is one of the tools used to evaluate this aspect of risk [4]. Risk perception refers
to the subjective experience of risk in potential traffic hazards or is considered
as a precursor of present driving behaviour [5]. Far from tackling every aspect
of this concept of risk, our research particularly sets out to understand how a
driver becomes aware of risk, and how he / she assesses the criticality of a
situation when this situation is likely to turn into an accident. [6].
This work is in line with the research carried out at INRETS (The French
national institute for transport and safety research) concerning drivers’ mental
activities modelling and simulation [7]. From a methodological point of view, this
study continues the experimental protocols developed and validated at the
LESCOT (Laboratory of Ergonomics and Cognitive Sciences applied to
Transport) for scientific purposes investigating drivers’ mental representations
[8], [9]. Indeed, the awareness of risk and the assessment of criticality are two
particular aspects of the Situation Awareness [10], mainly focusing on driving
situations presenting a risk of collision. Lastly, this research continues the
ARCOS project [11] dedicated to the design of “adaptive collision avoidance
devices”. Part of the ARCOS project has been devoted to analysing driving
287
Human Centred Design for Intelligent Transport Systems
activity when approaching fixed or dynamic obstacles. Within this framework,
this naturalistic experiment gathers data collected for twenty-two hours during
which 1200 kilometres were covered and over 110 situations with a risk of
collision were encountered [12]. All the situations presenting a risk have been
analysed. Then, a Driving Activity Model (DAM) in critical driving condition was
developed according to these data. When approaching a slow vehicle, the DAM
model distinguishes 6 main cognitivo-behavioral phases implemented by the
driver in order to avoid an accident [11]:
1. The normal driving situation phase, preceding the occurrence of the
obstacle.
2. The realization or risk awareness phase (i.e. the moment when the driver
detects the occurrence of an obstacle / the risk of collision and realizes that
the situation is critical).
3. The recovery phase (i.e. the driver acts on the situation to avoid collision
with the obstacle).
4. The stabilized driving phase, once the driver recovers the situation and
controls risks of collision with the obstacle but this does not necessarily
mean that the risks incurred are definitively eliminated.
5. The resolution phase, generally amounting to overtaking a slow vehicle
when approaching it.
6. Finally, the situation becomes normal again once the obstacle is really out
of view.
The focus point of this new research deals with the in-depth analysis of
cognitive mechanisms involved during the second phase of this DAM Model: the
risk awareness stage. We will thoroughly study the cognitive phase during
which the driver realizes that the driving situation, that was normal until then, is
suddenly becoming critical. The research aims are the human errors analysis
and the cognitive modelling of abilities implemented by the driver/rider to detect
or not the critical nature of a driving situation. To study risk awareness, we
established a protocol based on 21 ARCOS video sequences (filmed on board
a car) and presenting a risk of collision. These sequences are accompanied
with a Likert Scale, and with an Attitude Scale in the form of a semantic
differential based on Osgood’s work (e.g. [13]) in order to refine the quality of
the participants’ subjective assessments. On the base of this methods, it is then
possible to study and compare risk assessment performances of different
populations of drivers. In this paper, we will present the main results obtained
for two categories of drivers: one group of car drivers versus one group of
motorcyclists.
The reason why we selected a population of motorcyclists is that over the last 5
years, there has been a significant increase in the number of accidents
involving motorcycles on the European roads (+41%). According to the
European Road Safety Action Programme [14] 14% of motorcyclists were killed
in 2003 although the number of people killed on the road decreased by 12%
over this period. Thus, motorcyclists are particularly vulnerable and sensible of
risk wherever they are in Europe [15], [16]. In France, 807 motorcyclists were
288
Diversity and specificity of road user groups
killed and 17 390 were seriously injured on the road in 2006. Therefore,
motorcycle has rapidly become the most dangerous vehicle compared to all the
means of transport used on the road [17]. Besides, for each kilometre covered,
motorcyclists’ risk of being killed is 20 times higher compared to drivers of
standard four-door cars. This leads us to suggest that motorcyclists are forced
to show a greater vigilance as they are perfectly aware of their own vulnerability
among the other road-users. Although they comply to the same highway code
and they resort to the same cognitive processes as car drivers, their vehicle
remains more sensitive to weather conditions (rain, black ice…) and
infrastructures (slippery road surfaces marking on rainy days, holes or uneven
road surface, crash barriers that are dangerous when falling off a motorbike).
Moreover, they are more severely injured when they fall off their motorcycle as
they have no shell to protect them. The other drivers have more difficulty to
perceive them because their vehicle is smaller and faster. They are more fragile
within the road network. Nevertheless, according to a study carried out at the
French national institute for transport and safety research – INRETS [18]
dealing with two-wheelers crashes, motorcyclists have generally more
weaknesses during the prognostics and execution phases at the level of
cognitive functions while errors made by car drivers rather occur on perceiving
and diagnosing situations.
When considering the vulnerability of motorcyclists, they should be more
sensitive of risk and should anticipate the evolution of the traffic situation earlier
to start a regulation action if the situation worsens. From this fact, it is possible
that they do not perceive risks in the same way as car drivers. Certain situations
that are considered as not critical by car drivers may be considered as
dangerous by motorcyclists owing to a feeling of vulnerability. Conversely, it is
also possible that other situations may appear as critical by car drivers, whereas
they will be considered as not risky by motorcyclists. In some cases, these last
differences could explain the higher level of risk of accident for motorcyclists.
This research tries to examine these alternative hypotheses.
2 Methods
The methodology implemented is based on the presentation of 21 ARCOS
video sequences (filmed on board a car) of driving situations presenting a risk of
collision. These video sequences are divided into 6 categories of situations:
approaching fixed obstacles, approaching intersections, changing lanes,
following a vehicle, approaching slow vehicles, presence of vulnerable
obstacles (pedestrians and cyclists). The participants are asked to stop the
sequence when they feel that the situation becomes critical. Moreover, at the
end of the sequence, two measurement scales are then submitted to the
participants (see figure 1):
•
A Likert-type scale [18]: The participants assess the level of risk by moving
a cursor sliding along a scale with no graduation. The situation can thus be
quantified in terms of criticality on this scale, ranging from 0 % (not critical)
to 100 % (high level of criticality).
•
Attitude Scale in the form of a semantic differential based on Osgood’s work
[13], in order to refine the quality of the participants’ subjective
289
Human Centred Design for Intelligent Transport Systems
assessments. This differential is made of 16 antonyms defined for the
specific context of driving under critical situation. The use of a semantic
differential is complementary to an intensity scale as it helps us investigate
the different cognitive dimensions underlying the notion of criticality while
driving. This semantic differential is made of 4 dimensions: feelings,
predictability, description and implication. Each dimension has respectively
4 antonyms.
Fig.1. Synthetic view of the Risk Awareness Measurement Tool, including a
Likert Scale (on the left, below the road scene view) and an Osgood Semantic
Differential (on the right)
3 POPULATIONS
We recruited 21 participants altogether. We chose two population samples of
drivers in order to check if there were significant differences or not, in terms of
risk awareness, between car-drivers and motorcyclists. Owing to the exploratory
nature of this study, the size of our samples is limited. We should be cautious
about generalizing the results in terms of representativeness of all the French
motorcyclists versus car drivers. Nevertheless, we have built two homogenous
groups of riders / drivers, regarding their driving experience and age, to
increase the interest of the comparison.
Car-driver population: 11 experienced car-drivers (having a category B licence
for over 4 years and covering 5000 kilometres per year) participated to the
experiment. They have never driven a motorcycle before. They were all about
the same age, between 22 and 30 years old (mean age: 28).
Motorcyclist population: 10 motorcyclists participated to the experiment. They all
have a Motorcyclist “A” Licence and they rode their bike regularly, in terms of
frequency and distance (over 2000 kilometres/year by motorbike). They are
between 21 and 52 years old (mean age: 32). Some of them also had a car, but
their main means of transport was their motorcycle.
290
Diversity and specificity of road user groups
4 Main Results
The first experimental task of the participants was to quantify the level of
criticality by means of a Likert scale. This scale had no graduation, but the
results were subsequently reported in percentage (from 0 to 100 %). At this
stage, we then collected several results:
Overall criticality values (via the Likert scale): the criticality average value
assessed by all the “car drivers” participants for all the sequences is 57%,
against 34% for “motorcyclists” participants. This difference is statistically
significant (p< 0,001; T of student test). If we look into the detailed answers
given by the participants for each type of sequences, we notice that the
category considered as most critical (74%) by the car drivers corresponds to
vulnerable obstacles but the motorcyclists considered the same category almost
half as critical (38.5%). Moreover, the category considered as most critical by
the motorcyclists corresponds to intersections (45%). The car drivers also
considered these situations as critical (63%). For this category of situations, the
criticality differential for both populations is the smallest (18%). Finally, the
categories linked with the presence of an obstacle or depending on an
interaction with another vehicle (car, lorry or tractor) are on the whole
considered to be not very or not critical at all by motorcyclists (of the order of 20
%). On the other hand, the criticality level of each of these categories assessed
by car drivers remains around 50% (from 48% to 56%).
Table 1. Average value of criticality, according to categories of situations
Car drivers
Crit.
%
Motorcyclists
Crit
%
63
category 1 : approaching fixed
obstacles
category 2 : approaching
intersections
category 3 : changing lanes
59
category 3 : changing lanes
52
category 4 : following vehicles
48
23
34%
category 1 : approaching fixed obstacles
category 2 : approaching intersections
56
category 6 : presence of vulnerable
obstacles
74
category 4 : following vehicles
category 5 : approaching slow
vehicles
category 6 : presence of vulnerable
obstacles
Average value of criticality
58%
Average value of criticality
category 5 : approaching slow vehicles
48
27
45
21
38
Criticality values, according to categories of driving situations: We
classified our 21 video sequences in categories according to the type of
situations that may cause accidents such as approaching fixed obstacles,
approaching intersections, changing lanes, following vehicles, approaching slow
vehicles, presence of vulnerable obstacles (pedestrians and cyclists). Table 1
present results showing that, by and large, car drivers judged the situations as
more critical than motorcyclists. The main significant differences observed
between Motorcyclists and Car drivers concern categories 1, 4 and 6.
Categories of situations according to the averages of criticality
assessment given by the participants: If we classify the 21 video sequences
according the criticality assessments obtained by the two populations, we can
see four categories of situations (figure 2). The car drivers and the motorcyclists
291
Human Centred Design for Intelligent Transport Systems
feel that 7 sequences out of 21 are critical. They more particularly concern
approaches to vulnerable obstacles and one case of failure to give way. On the
other hand, 4 sequences are considered as not very or not critical at all by all
the participants. Moreover, 8 situations are assessed as critical by the car driver
but less critical by the motorcyclist. They more particularly concern approaches
to fixed obstacles. Finally, there remain two sequences that the motorcyclists
considered more critical in comparison with the car drivers proportionally to their
average. Indeed, these two sequences show situations involving a fixed
obstacle detected with difficulty and rather belatedly (a lorry parked at the end
of a curve) then a sequence filmed in bad weather showing a bus that rapidly
enters a roundabout.
2 situations
considered as
critical by the
motorcyclist but
less critical by
the car drivers
7 situations
considered as
critical by both
populations
8 situations
considered as
critical by the car
drivers but not
critical by the
motorcylists
4 situations
considered as not
critical by both
populations
Fig.2. Groups of situations, according to criticality averages among the two
populations
Early detection: The car drivers stopped the video sequences from 0.01 to
9.69 seconds earlier than motorcyclists did. However, the motorcyclists stopped
4 sequences out of 21 earlier (from 0.26 seconds to 2.86 seconds).
Number of sequence-stops: The car drivers tend to stop the sequences
several times (they stop the sequence on risk detection under driving situation).
Here is the total number of sequence-stops by all the subjects and of all the
sequences: 58 sequence-stops for the car drivers and 21 stops for
motorcyclists.
Results concerning the Osgood semantic differential: The participants also
had to complete the grid of the semantic differential based on Osgood’s
approach (table 2) and made of 16 antonyms classified according to 4
dimensions: (1) Predictability (e.g. probability to occur), (2) Feelings (e.g. to be
afraid or not), (3) Implication (e.g. to feel responsible or not of the criticality) and
(4) Driving situation characteristics (e.g. complexity level of the driving
situation). We noticed that all the participants - car drivers as well as
motorcyclists - considered the major part of our 21 sequences as Dangerous,
292
Diversity and specificity of road user groups
Fast and Abnormal. However, the car drivers judged these 3 antonyms
significantly far more strongly than riders (p< 0.05, T Student test). For instance,
the car drivers gave 67% to the antonym Fast, against 52% for the
motorcyclists. More interesting, we also noticed that the two groups clearly
disagree about two antonyms. Indeed, the motorcyclists generally considered
the sequences as Controllable, and they felt more Responsible for the situation.
Conversely, the car drivers considered more frequently the sequences as
Uncontrollable, and they felt less Responsible of the criticality of the driving
situation.
Table 2. Answers given by the participants to the Osgood semantic differential*
(Shaded items show a significant difference via Student Test: p<0.05)
Antonyms
Safe / Dangerous
Stressful / Calm
Disturbing / Reassuring
Difficult / Funny
Soliciting / Ordinary
Irresponsible / Responsible
Brought about / Suffered
Uncontrollable/Controllable
Probable / Improbable
Rare/ Frequent
Abnormal / Normal
Unpredictable / Predictable
Slow / Fast
Complex / Simple
Dynamic / Static
Forced / Open
Drivers (%)
71.80
33.08
31.69
67.52
44.12
39.25
57.44
45.21
56.65
73.62
35.35
56.72
67.47
51.51
38.52
64.74
Motorcyclists (%)
57.69
42.59
40.36
65.74
49.56
58.42
68.75
64.92
51.21
67.10
46.96
59.43
52.67
57.15
36.60
70.55
(*NB: note that for this Osgood’s scale, a value of 50% is the “origin” and means
that not any of the 2 antonyms is chosen. If % value is < 50%, the situation is
then defined by first antonym (like “Safe”). If % value is > 50%, it means that the
2nd antonym (like “Dangerous”) is assessed as more relevant to describe the
driving situation
Typical example of diverging answers between the car drivers and the
motorcyclists: To conclude this part dedicated to the results presentation, we
would like to present detailed data collected concerning a typical case of
diverging answers between the two populations (see Figure 3). The case
(Sequence n°21) corresponds to a situation assessed as very critical by the car
drivers (69%), but not critical at all by motorcyclists (18%).
Diverging answers between drivers and riders for this sequence could be
explained like that. During this situation, after having passed a bend in a
roadwork zone, the participants see a pedestrian finishing crossing the road (20
meters in front of our vehicle). For the car drivers, this situation is critical due to
293
Human Centred Design for Intelligent Transport Systems
the combining effect of the pedestrian presence on the left (after he crossed the
road) and the car incoming in the opposite direction. Consequently, they must
reduce their speed to avoid an accident. Conversely, it is easier for
motorcyclists to manage risk in this case: only the pedestrian’s crossing
behaviour could be potentially critical, but as he is now walking on the left side
along the road, the criticality is low for riders and it is possible to bypass him
without any interference with the incoming car on the opposite way. Semantic
differential results show that this situation is assessed as highly Dangerous
(83.91%), Fast (89.09 %), Stressful (29.45%), Uncontrollable (33.09 %),
Disturbing (21.27%), Abnormal (26.82%) and Irresponsible (17.27%) by car
drivers. On the contrary, this situation is identified as not Dangerous (54.5 %;
reminder: the “0” value for Osgood scales are 50%), but very Frequent (70%),
Predictable (70.70%), Simple (61.80%) and Controllable (59.30%) by
motorcyclists. These results clearly confirm the interest and the validity of our
“Risk Awareness Measurement” methods for studying and understanding the
nature of divergences between different populations of road users, concerning
risk assessment.
SEQUENCE n° 21:
Last view of the video sequence
ANTONYMS
Driver
Rider
Safe / Dangerous
83.91
54.50
Stressful / Calm
29.45
46.70
Disturbing / Reassuring
21.27
42.10
Difficult / Funny
66.82
57.10
Soliciting / Ordinary
36.00
43.90
Irresponsible/Responsible
17.27
53.00
Brought about / Suffered
38.82
48.80
Uncontrollable/Controllable 33.09
59.30
Probable / Improbable
54.82
54.70
Rare / Frequent
62.64
70.60
Abnormal / Normal
26.82
56.70
Unpredictable / Predictable 59.00
70.70
Slow / Fast
89.09
67.30
Complex / Simple
48.36
61.80
Dynamic / Static
36.64
35.20
Forced / Open
49.27
47.00
Osgood semantic differential results for
the each population (Shaded items show
a significant difference: p<0.05)
CAR DRIVERS
MOTORCYCLISTS
Criticality score for Car Drivers (Likert
Scale): 69%
Criticality score for Motorcyclists (Likert
Scale): 18%
Fig.3. detailed results collected for sequence n° 21
294
Diversity and specificity of road user groups
5 Discussion
The purpose of this research was to elaborate a methodology allowing us to
collect assessments of risk and to evaluate traffic situations presenting a risk of
collision among two populations of experienced drivers (11 car drivers and 10
motorcyclists). Motorcyclists are supposed to be more aware of risk as they are
very vulnerable on the road in case of accident. From this point of view, they
should judge “risky” situations as more critical and be aware of risks earlier.
As far as criticality assessment is concerned (grade given in percentage
between 0 and 100), the results show that, on average, the car drivers consider
the situations presented as more critical than the motorcyclists (reminder: 53%
for the car drivers and 34% for the motorcyclists). Moreover, the car drivers stop
the sequences more often and earlier on average. Nevertheless, we may say
that some of our driving situations represent a minimal risk for a motorcyclist
while they are perceived as risky by a car driver. For instance, we had 4 video
sequences showing approaches to fixed obstacles which are very easy to
manage for a motorcyclist but more difficult for a car driver. Consequently, the
average value of criticality for all the sequences is not the main relevant result in
this experiment. The most interesting results concern in-depth analysis
according to the type of traffic condition and driving context.
Indeed, a thorough analysis of our results shows that the risk assessments of
our video sequences between the two populations are not homogenous
according to our different categories of situations. There were 6 different
categories of situations. Approaching fixed obstacles or approaching slow
vehicles are not considered as critical by motorcyclists since it is easier for a
motorbike to bypass these obstacles when a car driver should slow down or
stop. On the other hand, approaching vulnerable obstacles (pedestrians,
bicycle, etc.), and more especially crossing intersections is on the whole
assessed as far more critical by the motorcyclists in comparison with their
overall criticality value of all the situations in so far as they fear that the car
drivers may not adapt their speed correctly. More simply, they fear that car
drivers may not spot them in time. Besides, accidentology data show that half of
the accidents involving a PTW and a car on an intersection are due to the car’s
failure to give way in 55% of the cases [19]. It is interesting to notice that
motorcyclists often considered our situations as critical when they took place in
unfavourable weather conditions. But generally, what is often considered
dangerous for motorcyclists is the quality of road surface, holes, gravels, rain
which is normal since the security of the motorbike is mostly ensured by the grip
of the machine on the road.
Concerning the semantic differential, it is interesting to mention that the
situations are judged as “controlled” and “responsible” (i.e. involving the rider /
driver’s responsibility) by motorcyclists, while it is the contrary for car drivers.
These 2 antonyms refer to the notion of involvement: the fact of feeling
“responsible” for what is happening and the fact of feeling able to “control” the
situation. Those last results are interesting. Indeed, being perfectly aware that
they are very vulnerable in case of accidents, this awareness tends to make
motorcyclists more aware of risks. That is what the differences observed for
these two antonyms seem to reveal.
295
Human Centred Design for Intelligent Transport Systems
6 Conclusion and perspectives
The general framework of this research was to analyse the cognitive
mechanisms involved in the assessment of collision risks when approaching
fixed or dynamic obstacles. We compared two populations of road users, one of
which being particularly exposed to risk (motorcyclists).
As the size of our driver samples is limited, we should be cautious about
generalizing the results in terms of representativeness of all the French
motorcyclists versus car drivers. Nevertheless, the protocol developed during
this experiment appeared to be very discriminating since it enabled us to collect
very different data between the two populations. We can also notice that
criticality assessment differences, observed between car drivers and
motorcyclists, differ or not according to the various categories of driving
situations.
Within the framework of a current research, we will question more car drivers
and motorcyclists. Concerning riders, more specifically, a new research is in
progress at LESCOT with the aim to compare risk awareness abilities for
different profiles of motorcyclists (e.g. novice versus experienced riders,
motorcyclists who like sport motorbikes, versus Harley Davidson or Scooters).
In this work, the characteristics of the riders’ personality are also assessed by
studying Zuckerman Sensation-Seeking Scale [20].
We will also draw a parallel between our video methods and other more classic
verbal methods implemented for Driver’s attitudes towards risk study and/or
human errors analysis (like the DBQ questionnaire; e.g. [2, 16]), with the aim to
evaluate correlations between social cognition on one hand, and risk awareness
abilities, on the other.
7 References
[1]
[2]
[3]
[4]
[5]
[6]
Reason, J. ‘Human Errors’, Cambridge University Press, 1990.
Musselwhite, C. ‘Attitudes towards vehicle driving behaviour: Categorising
and contextualising risk’, Accident Analysis and Prevention, 2006, 38, (2),
pp. 324-334
Turner, C., McClure, R., Pirozzo, S., ‘Injury and risk-taking behavior — a
systematic review’, Accident Analysis and Prevention, 2004, 36, (1), pp.
93-101
Jonah, B., Thiessen, R., Au-Yeunge, E, ‘Sensation seeking, risky driving
and behavioral adaptation’ Accident Analysis and Prevention, 2001, 33,
(5), pp. 679-684
Machin, M., Sankey, K., ‘Relationships between young drivers’ personality
characteristics, risk perceptions, and driving behaviour’ Accident Analysis
and Prevention, In Press, Corrected Proof, Available online 17 September
2007.
Banet, A.: ‘Définition d’une méthodologie pour l’étude du jugement de
criticité et de la conscience du risque de collision‘, Note de Recherche de
Master1, University of Lyon 2, 2005
296
Diversity and specificity of road user groups
[7]
[8]
[9]
[10]
[11]
[12]
[13]
[14]
[15]
[16]
[17]
[18]
[19]
[20]
Bellet, T., Bailly-Asuni, B., Mayenobe, P., Georgeon, O. ‘Cognitive
modelling and computational simulation of drivers mental activities’, in P.
Cacciabue and C. Re, Critical Issues in Advanced Automotive Systems
and Human-Centred Design, London, Springer Verlag, pp. 317-345.
Bailly, B.:‘Conscience de la Situation des conducteurs : aspects
fondamentaux, méthodes et applications pour la formation des
conducteurs’. PhD thesis, University of Lyon 2, 2004
Bailly, B., Bellet, T., Goupil, C.: ‘Driver’s Mental Representations:
experimental study and training perspectives‘. Proceedings of the First
International Conference on Driver Behaviour and Training, Stratford-uponAvon, England, Proceedings on CD-Rom, 8 p., 2003
Endsley, M. R.: ‘Toward a theory of situation awareness in dynamic
systems’. Human Factors, 1995, 37, (1), pp. 32–64
Bellet, T.: ’Driving activity analysis and modelling for adaptive collision
avoidance systems design’. Amsterdam: Elsevier. EA 2006 Congress (CDROM).
Bellet, T., Tattegrain-Veste, H., Deleurence, P., Goupil, C., Piechnik, B.,
Chanut, O. : ‘Analyse du contexte de conduite et du comportement du
conducteur pour une gestion adaptative de la coopération HommeMachine : Problématique, méthodologie et données recueillies‘ Rapport
ARCOS n° R030101-2/LESCOT, 45 p. 2003
Osgood, C., Suci, G., Tannenbaum, P.: ’The Measurement of meaning’,
University press of Illinois, 1967.
http://ec.europa.eu/transport/roadsafety_library/care/doc/annual_
statistics/2005_transport_mode.pdf
http://www.fema.ridersrights.org/docs/EAMS2007.pdf
Elliot, M.A., Baughan, C.J., Sexton, B.F.: ‘Errors and violations in relation
to motorcyclists’ crash risk’. Accident Analysis and Prevention, 2006, 39,
(3), pp. 491-499
Filou C, Lagache M, Chapelon J. Les motocyclettes et la sécurité routière
en France en 2003, ONSIR, la Documentation Française, Paris, 98 p.,
2005
Doise, W., Clemence, A., Lorenzi-Cioldi, F. : ‘Représentations sociales et
analyses de données’, Presses universitaires de Grenoble, 1992
Van Elslande, P.: ’Accident de deux-roues à moteur, vers une meilleure
compréhension grâce aux scénarios d’erreur’ Fiche de synthèse
Accidentologie et sécurité routière, 2002
Zuckerman, M., Kolin, E. A., Price, L.: ‘Development of a sensationseeking scale’, Journal of Consulting Psychology, 1964, 28, (6), pp. 477482
297
SESSION 6: DRIVERS’
ACCEPTANCE OF ASSISTANCE
FUNCTIONS
299
Drivers’ acceptance of assistance functions
DRIVERS’ RELIANCE ON LANE KEEPING
ASSISTANCE SYSTEMS: EFFECTS OF DIFFERENT
LEVELS OF ASSISTANCE
Anke Popken (Chemnitz University of Technology, Germany),
Lena Nilsson (VTI, Sweden),
Josef F. Krems (Chemnitz University of Technology, Germany)
anke.popken@phil.tu-chemnitz.de
ABSTRACT: Overreliance on Advanced Driver Assistance Systems (ADAS)
and a lack of involvement in the driving task reflect current safety-concerns in
connection with the increasing automation of the driving task due to ADAS. It is
hypothesised that those effects gain relevance the more ADAS intervene in the
cognitive and regulatory processes underlying driving. In a driving simulator
study drivers’ preparedness to divert attention away from the driving task was
investigated as a function of the level of assistance. Forty-five drivers drove 55
km on a simulated rural road with either one of three levels of lane keeping
assistance: (1) a high level of assistance realised by a Heading Control system,
(2) a low level of assistance realised by a Lane Departure Warning system, and
(3) no assistance (control condition). Drivers’ attention allocation to a visually
demanding secondary task served as a measure for their reliance on the
systems.
1 Introduction
In an attempt to increase road safety and to reduce the number of traffic deaths
and injuries, active safety in-vehicle technologies gain more and more
importance as a complement to passive safety technologies. The main aim is
not only to reduce actual crash consequences, but also to minimise crash risk.
Advanced Driver Assistance Systems (ADAS) are electronic in-vehicle systems
that offer support to the driver while driving and aim at enhancing driver (and
traffic) safety and comfort. ADAS intervene to different degrees in the driving
task and in vehicle control. Although ADAS are supposed to positively influence
driver behaviour and safety, the full range of possible adaptation effects in
response to the introduction of these system is at present difficult to predict
(e.g., [1]). Safety concerns with ADAS are primarily related to possible negative
human performance consequences that arise in conjunction with the automation
of task processes, as extensively studied in other domains like aviation ([2]).
Because ADAS - according to their type and functionality - intervene in different
levels of the control processes involved in driving, they change the way the
driving task is performed and affect the information processing and action
regulation processes involved in driving. The “relieving” function of ADAS may
indeed positively influence driver performance and comfort in non-critical
(“normal”) driving situations, but may have negative impacts in situations that
suddenly demand for drivers’ intervention; for instance situations that lie beyond
301
Human Centred Design for Intelligent Transport Systems
the ADAS’ functional scope or disturbing events such as system failures or
malfunctions.
In aviation research, evidence was found for problematic adaptation effects
such as overreliance on systems and inappropriate trust in system capabilities
([3, 4, 5, 6, 7, 8, 9]), complacency ([10, 11, 12]), vigilance problems and high
mental workload due to system monitoring demands ([13, 14, 15, 16, 17, 18]),
underload ([19, 20, 21]), and reduced Situation Awareness ([22, 23, 24, 25,
26]).
It is suggested that the occurrence and size of possible problematic adaptation
effects is related to the degree the operator is taken “out-of-the-loop” from the
controlled processes (e.g. [25, 27, 28]). Thus, system characteristics like the
design of the human-machine interaction and the systems’ performance (e.g.,
reliability) play an important role in the evolution of adaptation effects. A number
of studies demonstrated the strong relationship between system reliability and
its effects on human trust and monitoring behaviour ([29, 30, 4, 6, 7, 9, 10]).
People tend to rely on and to neglect monitoring of highly reliable systems, after
they have gained a sufficient level of trust in their performance. Highly reliable
systems reduce the necessity for human intervention in the automated
processes. A reduced involvement in the task, coupled with changes in
cognitive (information processing) and regulatory processes (e.g., vigilance) are
most likely consequences. Similar effects are predicted for the impact of the
level of automation, i.e. the level of control the operator has in the humanmachine interaction (e.g. [22]). In the automotive area, overreliance on ADAS
and a lack of involvement in the driving task (e.g., reduced Situation
Awareness, attention shifts to driving-unrelated activities) reflect major safetyconcerns in response to the introduction of ADAS. In this study drivers’
preparedness to divert attention away from the driving task is investigated as a
function of the level of (lane keeping) assistance. Thereby the focus laid on an
initial interaction with a system under high demanding conditions that were
designed to promote reliance on the system. Of special interest are potential
changes in drivers’ reliance after encountering critical situations where they
have to override the system or to ignore the system’s actions.
2 Methods
The VTI advanced moving base driving simulator III in Linköping, Sweden, with
a 120 degrees horizontal field of view was used in this study (Figure 1).
Participants drove six times on a rural road of 11 km length with one driving
lane in each direction. The lane width was 3,5 m. The speed limit during the
study was 70 km/h.
302
Drivers’ acceptance of assistance functions
Fig. 1. The VTI advanced driving simulator III.
2.1
Participants
Forty-five participants (23 male, 22 female) aged between 25 and 45 years (M =
33,7; SD = 6,4) took part in this study. They had been driving at least 5000 km
during the last year. Most of them had been driving in the VTI driving simulator
before.
2.2
Levels of lane keeping assistance
The 45 participants were randomly assigned to one of three experimental
conditions (15 participants in each condition) representing three different levels
of assistance in lane keeping: (1) Driving with a Heading Control system (HC
condition) corresponding to the highest level of lane keeping assistance, (2)
driving with a Lane Departure Warning system (LDW condition) corresponding
to a low level of lane keeping assistance, and (3) driving manually (MD
condition) without any assistance in lane keeping (control condition). The LDW
system warned the driver via a vibration in the steering wheel when he or she
crossed the lane markings. The HC system guided the drivers within the lane by
applying counter forces on the steering wheel as soon as they were
approaching the left or right lane marking. Both systems were 100% reliable
during the whole study.
2.3
Secondary Task
Concurrently to driving participants performed a visually demanding secondary
task: The Arrows Task developed in the European HASTE project. The visual
distraction caused by this task should provoke lateral path deviations and thus
enable drivers to experience the actions of the lane keeping assistance systems
and to develop trust in the system. Furthermore, drivers’ attention allocation to
the primary driving task and the secondary task was used as an indicator of
their reliance on the assistance systems (compared to non-assisted driving).
The Arrows Task was presented on a 7 inches TFT touch screen mounted in
the centre console of the simulator vehicle. The secondary task display was
positioned on a horizontal eccentricity of about 22 to 37 degrees visual angle to
the left of the driver’s straight-ahead line of sight and on a vertical eccentricity of
about 13 to 22 degrees visual angle below the driver’s straight-ahead line of
sight. The Arrows Task started on certain road marks which were the same for
303
Human Centred Design for Intelligent Transport Systems
each driver in each condition. The duration of one Arrows Task was 30 s. One
Arrows Task consisted of a series of displays presenting different configurations
of 16 arrows pointing in different directions. For each display drivers had to
decide if there was an arrow present pointing upward and to accordingly press
the “yes” or “no” buttons in the upper or lower part of the screen, respectively.
The time out for each display was 5 s. Response times above 5 s were
recorded as missing responses. Each presentation of a new display was
signalled by a sound. Additionally, drivers got an auditory and visual
performance feedback.
2.4
Procedure
Participants were instructed to drive as they would usually drive on a real road
with corresponding character concerning interaction with other traffic
participants, speed choice, manoeuvring, and time gaps. Participants were
instructed to use the turn indicator to signal their intention to depart from the
driving lane, e.g. during lane change manoeuvres. They were informed that the
activation of the indicator would suppress the lane departure warning and
deactivate the HC system’s steering torque. Participants in the HC condition
were additionally told that they could principally override the system if they
applied a counter force to the steering wheel greater than the maximum
steering force of the HC system.
Participants started with a training session in order to become accustomed to
driving with the simulator and with the LDW and HC system. During the training
session participants were instructed to perform some lane change manoeuvres
in order to experience the systems’ functionality. After that participants practised
the secondary task while the simulator vehicle was parked. Participants were
instructed to perform the Arrows Task as well and as fast as possible while
driving, but without jeopardising traffic safety.
Subsequently participants performed two experimental drives. During the first
drive (about 15 min.) participants were supposed to build up a stable level of
trust in the system and adapt their time-sharing strategies accordingly. During
the second drive (about 25 min.) drivers encountered eight critical situations in
which they had to react accordingly in order to not compromise traffic safety or
violate the traffic rules. The LDW and the HC system were 100% reliable during
the whole study (no system failures occurred). However, the critical situations
lay partly outside the systems’ functional limits and thus, drivers had to override
the systems. Of special interest are the potential changes in drivers’ trust,
reliance, and workload as a function of their experience of these critical
situations.
2.5
Dependent Measures
Performance-based measures of drivers’ time-sharing strategies (eye glance
behaviour, secondary task performance) were used as indicators of their
reliance on the lane keeping assistance systems. Additionally various subjective
measures were collected after the two experimental sessions in order to assess
amongst other things drivers’ alertness and mental workload, their perception of
304
Drivers’ acceptance of assistance functions
system performance, their trust in the systems, and their opinions about how
such systems may influence their safety and driving behaviour.
3 Results
All measures were subject to a 3 x 3 factorial repeated measures ANOVA with
the level of assistance (Heading Control, Lane Departure Warning, No
Assistance) treated as between-subjects factor and the session/criticality of
driving situation (session one – no critical driving situations, session two – no
critical driving situations, session two – critical driving situations) treated as
within-subjects factor. Some first results of the analysis are presented below.
3.1
Secondary Task Performance
Drivers performed 19 Arrows Tasks in the first experimental session and 25
Arrows Tasks in the second experimental session.
3.1.1 Response Accuracy
Response accuracy was generally very high for all experimental conditions.
Unassisted drivers had a higher percentage of correct answers and a lower
percentage of false and missed responses in the Arrows Task than LDW drivers
and HC drivers in both experimental sessions. This difference was however not
significant. The performance of the unassisted drivers improved from session
one to session two (in terms of percentage of correct and false responses),
whereas the performance of the LDW and HC drivers deteriorated from session
one to session two. The interaction between level of assistance and
session/criticality of driving situation was not significant. HC drivers generally
missed more responses than LDW and unassisted drivers (n.s.). There was a
highly significant effect (p < 0.001) for the occurrence of critical driving
situations, in that response accuracy deteriorated considerably in critical
situations compared to non-critical driving situations in both sessions.
3.1.2 Mean Number of responded displays
Each Arrows Task lasted for about 30 s. During one Arrows Task a number of
displays representing different arrows configurations were shown. Dependent
on how fast drivers responded to each single display, more or less displays
were presented during the whole duration of one Arrows Task. Drivers in the
HC condition on the average answered the smallest number of displays and
LDW drivers answered the largest number of displays during one Arrows Task
in non-critical driving situations. For HC and LDW drivers, performance in
critical situations in session two was better than performance in non-critical
situations in session one; whereas the performance of unassisted drivers was
worse during critical situations in session two compared to non-critical situations
in session one. There was neither a significant main effect for the level of
assistance nor a significant interaction between the level of assistance and the
number of session/criticality of driving situation. There was a significant
performance improvement for drivers in all conditions in non-critical driving
situations from session one to session two (p < 0.001).
305
Human Centred Design for Intelligent Transport Systems
3.1.3 Reaction time
Reaction times were calculated for all displays except those drivers missed to
respond to within five seconds (recorded as missing responses). LDW drivers
and drivers in the no assistance condition reacted faster than HC drivers in noncritical driving situations. This effect was not statistically significant. There was a
significant main effect for the session/criticality of driving situation factor, in that
reaction times were significantly longer during critical situations compared to
non-critical situations in both sessions for all levels of assistance (p < 0.001).
Reaction times decreased from session one to session two but this effect was
not significant.
3.2
Eye glance behaviour
This analysis is based on a total number of more than 13000 display glances.
3.2.1 Mean single glance durations to the secondary task display
HC drivers had the longest mean single glance durations to the display in
session one and the shortest mean single glance durations in session two.
Mean single glance durations to the display decreased from session one to
session two, but this difference did not yield statistical significance in post-hoc
analyses. However, there was a significant effect for the criticality of the driving
situation, in that mean single glance durations to the display were significantly
shorter during critical situations compared to non-critical situations (p < 0.001).
During critical situations, unassisted drivers had the longest display glance
durations compared to HC and LDW drivers.
3.2.2 Mean display glance frequencies
LDW drivers had the highest frequency of glances to the display; and HC
drivers had the lowest display glance frequency in session one. During session
two, HC driver looked less often to the Arrows Task display than LDW and
unassisted drivers. Unassisted drivers looked most often to the display in
session two. There was no significant effect for the level of assistance and no
significant interaction between the level of assistance and session number/
criticality of situation.
There was however a highly significant main effect (p < 0.001) for the session
number/ criticality of situation factor: Drivers in all conditions looked significantly
less often to the display in session two (compared to session one), and looked
least often to the display during critical driving situations.
4 Discussion
The results show that drivers adapted their task related effort to driving task
demands. In critical driving situations, drivers allocated fewer resources to the
secondary task, confirmed by a significantly worse secondary task performance
and significantly shorter and fewer glances to the secondary task display in
critical driving situations.
There is also evidence for a general training effect for drivers’ performance in
the secondary task. Drivers answered more displays in session two than in
session one, and reaction times slightly improved. Furthermore, drivers needed
306
Drivers’ acceptance of assistance functions
significantly fewer glances to extract the relevant information from the
secondary task display in session two compared to session one.
The results for the different levels of assistance show a more complex picture.
There was no significant main effect of the level of assistance for any of the
measures.
For the Arrows Task performance, there was a surprising finding in that
unassisted drivers performed better than HC and LDW drivers in terms of
response accuracy. This difference was also obvious in critical driving
situations. A further interesting finding was that HC and LDW drivers’ Arrows
Task performance deteriorated from session one to session two (in terms of
percentage of correct and false responses), whereas performance of unassisted
drivers improved. HC drivers had the highest number of missing responses
during both sessions and especially in critical driving situations. In critical driving
situations there seems to be a tendency for LDW drivers of tempting to respond
to displays rather than to omit them (even when the answer is false), whereas
HC drivers rather drop the secondary task. One possible explanation for these
findings could be that HC drivers had to invest some additional resources in
system control, which was particularly necessary in session two when they had
to recognise critical driving situations which required them partly to override the
HC system.
There seems to be some further evidence for interpretation in the results of the
eye glance measures. For HC drivers both duration and frequency of glances to
the secondary task display decreased from session one to session two. HC had
the longest display glance durations in session one, and the shortest display
glance durations in session two. For LDW and unassisted drivers there was
mainly a difference in glance frequencies from session one to session two (less
frequent display glances in session two), whereas display glance durations
were similar in both sessions. Unassisted drivers had the longest and also the
most frequent glances to the display in critical driving situations.
Further analysis of other glance behaviour measures (e.g. the percentage of
time spent looking to the display during one Arrows Task) and the relation of the
behavioural measures with the subjective data (e.g. mental workload) will
complete the picture of drivers resource allocation strategies when driving with
different levels of lane keeping assistance.
5 Acknowledgements
This study was funded by a grant for infrastructure-sharing from the HUMANIST
Network of excellence (HUMAN centred design for Information Society
Technologies; www.noehumanist.org). We would like to thank the following
persons at VTI for their overwhelming support in conducting this study: Anne
Bolling (for a lot of fruitful discussions and the timely management of the study),
Beatrice Söderström, Birgitta Thorslund and Jerker Sundström (for being
experimental leaders), Hǻkan Jansson and Laban Källgren (for realising the
simulator scenarios), Göran Palmkvist (for implementing the lane keeping
assistance systems), Mikael Adlers (for implementing the secondary task), and
307
Human Centred Design for Intelligent Transport Systems
Jan Törnros (for translating the always newest version of the questionnaires
and instructions into swedish).
6 References
[1]
[2]
[3]
[4]
[5]
[6]
[7]
[8]
[9]
[10]
[11]
[12]
[13]
[14]
Saad, F., Hjälmdahl, M., Cañas, J., Alonso, M., Garayo, P., Macchi, L.,
Nathan, F., Ojeda, L., Papakostopoulos, V., Panou, M., Bekiaris, E.: ‘AIDE
Deliverable 1_2_1: Literature review of behavioural effects’ (Subproject
SP1, Workpackage 1.2), 2004, Available at: http://www.aideeu.org/pdf/aide_d1-2-1final.pdf
Parasuraman, R., and Mouloua, M.: ‘Automation and human performance:
Theory and applications’ (Lawrence Erlbaum Associates, 1996)
Muir, B. M.: ‘Trust in automation: Part I. Theoretical issues in the study of
trust and human intervention in automated systems’, Ergonomics, 1994,
37(11), pp. 1905-1922
Muir, B. M. and Moray, N.: ‘Trust in automation. Part II. Experimental
studies of trust and human intervention in a process control simulation’,
Ergonomics, 1996, 39(3), pp. 429-460
Lee, J. D., and See, K. A.: ‘Trust in automation: Designing for appropriate
reliance’, Human Factors, 2004, 46(1), pp. 50-80
Lee, J. D., and Moray, N.: ‘Trust, self-confidence, and operators'
adaptation to automation’, International Journal of Human-Computer
Studies, 1994, 40, pp. 153-184
Lee, J. D., & Moray, N.: ‘Trust, control strategies and allocation of function
in human-machine systems’, Ergonomics, 1992, 35, pp. 1243-1270
Parasuraman, R. and Riley, V.: ‘Humans and automation: Use, misuse,
disuse, abuse’, Human Factors, 1997, 39(2), pp. 230-253
Dzindolet, M. T., Peterson, S. A., Pomranky, R. A., Pierce, L. G., and
Beck, H. P.: ‘The role of trust in automation reliance’, International Journal
of Human-Computer Studies, 2003, 58, pp. 697-718
Parasuraman, R., Molloy, R., and Singh, I. L.: ‘Performance consequences
of automation-included "complacency"’, The International Journal of
Aviation Psychology, 1993, 3(1), pp. 1-23
Moray, N., and Inagaki, T.: ‘Attention and complacency’, Theoretical Issues
in Ergonomics Science, 2000, 1(4), pp. 354-365
Farrell, S., and Lewandowsky, S.: ‘A connectionist model of complacency
and adaptive recovery under automation’, Journal of Experimental
Psychology: Learning, Memory, and Cognition, 2000, 26(2), pp. 395-410
Parasuraman, R., Molloy, R., Mouloua, M., and Hilburn, B.: ‘Monitoring of
automated systems’, in Parasuraman, R. and Mouloua, M. (Eds.):
‘Automation and human performance: Theory and applications’ (Lawrence
Erlbaum Associates, 1996), pp. 91-115
Warm, J. S., Dember, W. N., and Hancock, P. A.: ‘Vigilance and workload
in automated systems’, in Parasuraman, R. and Mouloua, M. (Eds.):
‘Automation and human performance: Theory and applications’ (Lawrence
Erlbaum Associates, 1996), pp. 183-200
308
Drivers’ acceptance of assistance functions
[15] Molloy, R., and Parasuraman, R.: ‘Monitoring an automated system for a
single failure: Vigilance and task complexity effects’, Human Factors, 1996,
38(2), pp. 311-322
[16] Hancock, P. A., and Warm, J. S.: ‘A dynamic model of stress and
sustained attention’, Human Factors, 1989, 31(5), pp. 519-537
[17] Desmond, P. A., and Hancock, P. A.: ‘Active and passive fatigue states’, in
Hancock, P. A. and Desmond, P. A. (Eds.): ‘Stress, workload, and fatigue’
(Lawrence Erlbaum Associates, 2001), pp. 455-465
[18] Parasuraman, R.: ‘Human-Computer monitoring’, Human Factors, 1987,
29(6), pp. 695-706
[19] Matthews, G., and Desmond, P. A.: ‘Task-induced fatigue states and
simulated driving performance’, The Quarterly Journal of Experimental
Psychology, 2002, 55A (2), pp. 659–686
[20] Desmond, P. A., Hancock, P. A., and Monette, J. L.: ‘Fatigue and
automation-induced impairments in simulated driving performance’,
Transportation Research Record, 1998, 1628, pp. 8-14
[21] Carsten, O.M.J., and Nilsson, L.: ‘Safety Assessment of Driver Assistance
Systems’, European Journal of Transport and Infrastructure Research,
2001, 1(3), pp. 225-243
[22] Endsley, M. R.: ‘Automation and situation awareness’, in Parasuraman, R.
and Mouloua, M. (Eds.): ‘Automation and human performance: Theory and
applications’ (Lawrence Erlbaum Associates, 1996), pp. 163-181
[23] Kaber, D. B., and Endsley, M. R., ‘The effects of level of automation on
human performance, situation awareness and workload in a dynamic
control task’, Theoretical Issues in Ergonomics Science, 2004, 5(2), pp.
113-153
[24] Endsley, M. R., and Kaber, D. B., ‘Level of automation effects on
performance, situation awareness and workload in a dynamic control task’,
Ergonomics, 1999, 42(3), pp. 462-492
[25] Endsley, M. E., and Kiris, E. O.: ‘The out-of-the-loop performance problem
and level of control in automation’, Human Factors, 1995, 37(2), pp. 381394
[26] Adams, M. J., Tenney, Y. J., and Pew, R. W.: ‘Situation Awareness and
the cognitive management of complex systems’, Human Factors, 1995,
37(1), pp. 85-104
[27] Buld, S., and Krüger, H.-P.: ‘Wirkungen von Assistenz und Automation auf
Fahrerzustand und Fahrsicherheit. Projekt: EMPHASIS (EffortManagement und Performance-Handling in sicherheitsrelevanten
Situationen, 19 S 9812 7, Bundesministerium für Bildung und Forschung),
Würzburg: Interdisziplinäres Zentrum für Verkehrswissenschaften an der
Universität Würzburg, 2002
[28] Parasuraman, R., Sheridan, T. B., and Wickens, C. D.: ‘A model of types
and levels of human interaction with automation’, IEEE Transactions on
Systems, Man, and Cybernetics - Part A: Systems and Humans, 2000,
30(3), pp. 286-297
309
Human Centred Design for Intelligent Transport Systems
[29] Wiegmann, D. A., Rich, A., & Zhang, H.: ‘Automated diagnostic aids: The
effects of aid reliability on users’ trust and reliance’, Theoretical Issues in
Ergonomics Science, 2001, 2(4), pp. 352-367
[30] Moray, N., Inagaki, T., & Itoh, M.: ‘Adaptive automation, trust, and selfconfidence in fault management of time-critical tasks’, Journal of
Experimental Psychology: Applied, 2000, 6(1), pp. 44-59
310
Drivers’ acceptance of assistance functions
USER ACCEPTANCE AND EFFECTIVENESS OF
WARNING AND MOTOR PRIMING ASSISTANCE
DEVICES IN CAR DRIVING
M. El Jaafari (Renault, France), J-F. Forzy (Renault, France),
J. Navarro (IRCCyN, France), F. Mars (IRCCyN, France),
J-M. Hoc (IRCCyN, France)
myriam.el-jaafari@renault.com
ABSTRACT: This paper deals with driving assistances that intervene when a
lane departure is imminent. Previous work [1] showed that motor priming
devices (devices prompting the driver to take action by means of an asymmetric
steering wheel vibration) were the most effective warning systems. The present
experiment tries to qualify the different lane departure warning systems not only
in terms of effectiveness but also in terms of acceptability. For this, both
objective and subjective data were collected. Firstly, the study shows that
acceptability and effectiveness are not necessarily correlated, and can even be
contradictory. From an ergonomics point of view, we claim that a compromise
should be reached between those criteria. Secondly, we show that a combined
mode (sound and motor priming) could be a solution to find such a compromise.
Finally, a perspective for further research is to better combine the modalities,
through improved synchronization of the driver sensations.
1 INTRODUCTION
In order to contribute to the ergonomic design of safety devices for critical
situations in cars, a research project on Human-Machine cooperation (PreditPrevensor) deals with lateral control situations where additional devices are
assumed to enhance safety. For Hoc and Blosseville (2003), safety devices in
car driving can be classified into four levels of Human-Machine cooperation [2]:
(1) perceptive mode, (2) mutual control mode, (3) function delegation mode and
(4) fully automatic mode. In the particular case of lateral control, some proposed
safety devices belong to mutual control mode. This means that drivers have full
control of their vehicles, but their driving behavior can be criticized (warning
mode devices). Moreover, safety devices can lead drivers to correct their
trajectory (motor priming) by inducing asymmetric vibrations in the steering
wheel for example.
Previous studies on safety devices applied to lateral control show ambivalent
results. The effect of motor priming in steering wheel was found to depend on
inter-individual variations ([3]; [4]), and in some cases, driver response went
opposite to the suggested action [3]. However, [1] demonstrated by a
comparative evaluation technique that, among five driving assistance devices,
all belonging to mutual control mode, motor priming was found to be superior in
terms of effectiveness. Indeed, the asymmetric vibrations on the steering wheel
seem to guide the driver to take the correct action. Nevertheless, acceptability
of such vibrations was not discussed.
311
Human Centred Design for Intelligent Transport Systems
In accordance with this finding, the present study aims at checking advantages
of the mutual control mode for lane departure prevention, and particularly the
superiority of motor priming when induced through the steering wheel. Beyond
effectiveness measures and theoretical utility [5], the users point of view
regarding similar driving assistances is assessed. Investigating drivers’
subjective side is particularly justified by the fact that Human-Machine
cooperation is considered as a management of interferences between different
agents [6]. Actually, the study of the representations of the system, the attitudes
and the motivations are crucial to understand how and why the human feels
either comfortable or reluctant to cooperate with the machine, in order to control
the dynamic driving situation. In other words, what the users think and expect
from the device, as well as how they perceive it, needs to be evidenced. More
generally, comfort or discomfort, pleasure or displeasure, satisfaction or
dissatisfaction ([7]; [8]) constitutes the heart of this framework.
In this paper, motor priming through the steering wheel is presumed to be more
effective than a warning, because the direct intervention on the action level
should improve corrective maneuvers and situation diagnosis. However, for the
same reasons, drivers could reject this assistance. Besides, a study on social
acceptability of driving assistance devices showed that users’ judgments are
unfavorable when the assistance takes more or less control of the driving
situation [9].
Following this idea, a mode that combines motor priming and auditory warning
is proposed. In this case, it is assumed that a sound reminding driving on
rumble strips is potentially acceptable. Furthermore, [10] showed that when
preexisting or direct association between the sound and the event is strong, this
pairing tends to be learned and retained more readily. This may induce that
auditory warning implying strong or direct association with the event should be
more understandable and thus acceptable. The aim of this paper is therefore to
check whether combining motor priming with auditory warning improves driver
acceptability along with keeping device effectiveness.
2 METHOD
2.1
Participants
Twenty participants (16 men and 4 women) ranging from 23 to 52 years old
(mean = 34) took part in this study. Sixty percent of them cumulated more than
10-years driving experience. Participants were volunteers from Renault workers
and none of them was involved in the design process of driving assistances.
2.2
Material
2.2.1 Simulation
Experimentations took place on a dynamic driving simulator (Cards2, developed
by the Technical Centre of Simulation of Renault). The platform consists of a
driver's cabin (on top of an hexapod) with a manual gearbox and a 150° vision
of the driving scene. The cockpit was specifically equipped with the following
devices: (1) haptic seat, (2) haptic steering wheel (vibration or asymmetric
312
Drivers’ acceptance of assistance functions
oscillations) and (3) 3D sound. Simulated driving scene was a country road,
where “legal” speed limitations were fixed at 80km/h on straight lines and
70km/h in curves.
2.2.2 Interview
An open interview inspired by explicitation interview techniques ([11]; [12]) was
performed. These techniques aim to explore implicit, pre-thought-out aspects of
a physical or mental action. The objective of the interview was to collect the
description of the action (effectively) realized by the driver. Interviewer has to
guide drivers to operate a thinking-out of their experience and to put it into
words. Techniques of explicitation allow « to help the user to describe his
activity such as it took place in this situation, such as what he did, perceived,
thought and felt during that specific situation» ([7]; [8]). Although we did not
perform a real explicitation interview as defined by Vermersch, we were strongly
inspired by those techniques by using non-inductive questioning. The aim was
to collect verbal reports on feelings, senses, internal states, thoughts that were
lived by the driver with a driving assistance device.
2.3
Driving assistances
Five driving assistance devices were assessed in this study. All these devices
were activated in case of imminent lane departure (with axial line or bank line).
¾ The auditory warning mode consisted of a lateral 3D sound coming
from the side where lane departure occurred. The emitted sound was similar to
a rumble strip noise.
¾ The wheel vibratory warning mode was produced by a lateral haptic
vibration from the side where lane departure occurred.
¾ The motor priming mode was generated by an asymmetric haptic
stimulation on the steering wheel. In other words, jolts on the steering wheel
indicated the direction of the trajectory to adopt. The stimulation was not
sufficient in itself to correct the trajectory.
¾ The seat vibratory warning was produced by a lateral haptic stimulation
from the side where lane departure occurred.
¾ The combined mode consisted of a combination between the auditory
warning mode and the motor priming mode.
2.4
Procedure
After a familiarization phase with the driving simulator, participants performed
ten laps, of five minutes each. Runs were alternated between lap, with and
without assistance. After assisted laps, the experimenter interrogated briefly the
driver about what he or she had just experienced. To control any order effect,
presentation of various assistances was counterbalanced among the
participants.
Driving scenarios were defined to control lanes departures. Every assisted ride
consisted of four critical events, two before the curves’ entry and two on
straights lines. The non-assisted ride also consisted of two to four critical
313
Human Centred Design for Intelligent Transport Systems
events. For each critical event, a slight gusts of wind occurred (drivers were
unaware of this occurrence) in parallel with a distraction task to provoke lane
departures. This relatively “ecological” secondary task was supported by a
screen placed in the driver's cockpit at the usual place of the radio and was set
up in order to distract the driver by means of a “reading task” (scrolling of a list
of three words per second).
Drivers had been instructed to adopt a "laid-back" driving attitude and to keep
both hands on the steering wheel in “10:10” or “9:15” time position. They were
informed that the beginning of the reading task would be randomly indicated by
a beep sound for several times. The importance to maintain attention on
reading while maintaining steering wheel angle was strongly emphasized until
the end of secondary task (which was synchronized with the assistance onset if
present). At last, drivers were invited to maintain speed driving as far as
possible with speed limit standard and to respect those standards.
Experimenter reassured participants that these experimental conditions were
made to provoke lanes departures.
After each lap, participants were asked to describe their own personal
experiences when lane departures occurred: what happened for them? What
were their feelings or thoughts? At the end of the experiment these short
descriptions were re-examined with the subject in a way to avoid inductive
questioning. These post interviews took place in the simulator’s cockpit to
facilitate drivers’ remembrance of the assisted driving. Afterwards, subjects
were invited to classify the assessed modalities in order of preference (without
ex-aequo).
2.5
Data analysis
Three types of data were analyzed: performance measures (lane departure
durations), rankings and verbalizations. Only rankings and verbalizations are
detailed in this paper for eighteen participants out of the twenty original because
of a destroyed tape. Further details about performance values can be found in
Navarro et al [13].
2.5.1 Ranking
Assistance devices were statistically analyzed in terms of order of preferences.
Ranking results were used to find which modality was the favorite and which
one was unfavorable. In addition, Friedman non-parametric test was used to
highlight, if applicable, significant differences between modalities.
2.5.2 Analysis of contents
The detailed analysis of verbalizations underlined four discursive categories: a)
Sensory sensation (pleasantness, inconvenience, unpleasantness, etc.); b)
Interpretation of the type of alert (lane departure), of its location (side
departure), and the corrective action required (towards the lane centre); c)
Perceived utility; d) Attitudes (satisfaction, dissatisfaction, etc.). Those
categories were evidenced thanks to content analysis methods. First, we
categorized participants’ speech in positive and negative dimensions; then we
extracted and classified spontaneous verbal reports able to describe drivers’
acceptability. In addition, the previously mentioned emergent categories are
314
Drivers’ acceptance of assistance functions
close to some well-known acceptability criteria in Human-Machine Interaction
field ([5]; [14] for instance). Through this analysis we wanted to identify in
details, from actual experiences of the subject, the inherent reasons for
approving or rejecting such or such modality.
2.5.3 Driving performances
The variable retained was the lane departure durations. Values were interpreted
in terms of effect of the assistance device on lateral departure. In other words,
we were interested in the time saved thanks to the assistance device.
3 RESULTS
3.1
Rankings
The rankings (fig.1) show a first approach of the compared levels of systems
acceptability.
Fig.1. Proportion by rank of the relative classification in order of preference
Figure 1 illustrates participants’ relative preferences for auditory warning. It
appears mainly as a second choice (50% in rank 2). While sound is globally
preferred, results regarding vibratory warnings (seat and steering wheel) are
scattered. In addition, combined mode is rather scattered (39 % in rank 1 and 2
and 34 % in rank 4 and 5). Finally, motor priming device obtained the worst
results, mainly classified between rank 4 and 5. Friedman test did not reveal
significant global differences of driver’s assessment of the modalities, probably
due to the few number of participants. Moreover, figure 1 shows that differences
between auditory warning and motor priming are sensible (tendency).
Furthermore, verbal reports (§ 3.2) corroborate this idea.
3.2
Analysis of content
¾ Auditory warning appeared to be the least intrusive modality. Indeed,
some positive comments underline the non-aggressive aspect of this modality:
“this one is really soft”; “this one is probably the least violent”… Analogy of the
simulated noise along with perceptible noise of real driving (rumble strips)
enabled drivers to represent themselves as having a lane departure. Attitudes
were rather favorable, but many drivers had some doubts about discriminating
315
Human Centred Design for Intelligent Transport Systems
this sound in real vehicle’s sound environment: “I thought there were a lot of
signals in the vehicle, so how can I distinguish this one from another one”…
¾ Motor priming device was outlined as the most intrusive modality.
Globally lateralization was generally not well perceived: “the jolts are not
indicative, it’s like a back and forth movement”... In addition, jolts in the steering
wheel disturbed some drivers who confused it with their usual driving
sensations on the steering wheel. This confusion could have caused troubles
such as a feeling of trajectory lost of control for example: “I have the impression
not to control the car”…
¾ Combined device presented different perceptive representations. Two
perceptive profiles were evidenced: those who used both signals and those who
used one of the most salient signals according to the individual’s sensibility: “I
had the feeling to react because of the sound and not because of the sensation
detected by my hands” or “the sound, it is like I did not hear it because there is
a lot of things”. Additionally, risks of confusion inherent to stimulation in the
steering wheel were evoked.
¾ Vibration warning on the steering wheel caused various
representations. First, lateralization was not well perceived (felt on both hands)
but some drivers reported that the signal allowed them to anticipate correction.
Besides, many participants were confused and annoyed.
¾ Vibration warning on the seat revealed a scattering of preferences.
Globally, lateralization was perceived without ambiguity but it was not
systematically associated to lane departure side. Furthermore, the location of
the vibrations can be annoying.
3.3
Acceptability vs. effectiveness
Auditory warning was generally preferred, without being ideal as reported by
some drivers. In addition, drivers rejected motor priming device whereas
opinions relative to combined device were scattered. Figure 2 shows driving
effectiveness with respect to tested modalities.
Level of acceptability
(Ranking)
Straight lines
Curves
Auditory warning
High
Seat (Vibratory
Warning)
Steering wheel
(Vibratory Warning)
Intermediate
Combine ((Auditory&
Motor Priming)
Motor Priming
Duration of lateral lane departure (s)
--0,8
-0,7
-
-0,6
-0,5
-
-0,4
-0,3
-
-0,2
-0,1
-
Low
0
Fig.2. Effects of the auditory warning [high acceptability] and the motor priming
[low acceptability] on lateral lane departure durations (normalized against
control condition)
For further details regarding performance measurements, see Navarro et al
[13].
316
Drivers’ acceptance of assistance functions
Statistical analysis revealed that motor priming device had a significant effect on
lane departure duration compared to control condition in straight lines as well as
in curves. Nevertheless, the majority of drivers significantly rejected this device.
The auditory warning also tends to decrease lane departure durations but not
significantly. Furthermore, compared with other assistance devices,
performance (duration of lateral excursions) was among the worst, although
paradoxically, it was significantly preferred than motor priming. Combined mode
allowed driver to be more successful with respect to other devices while
subjective judgments were scattered. Moreover, effectiveness and preferences
of vibratory warning on seat as well as on steering wheel presented scattered
results and thus were hardly distinguishable.
4 DISCUSSION
In this paragraph, the results are interpreted with respect to the hypothesis and
the general aim of this study. More specifically, integration of an acceptable and
effective driving assistance device for safety applied to lateral control is
discussed.
First, the experimental protocol was quite ecological and therefore relatively
close to a real driving situation, thanks to methodological choices. Dynamic
driving simulation has the particularity to immerse the driver in driving situation
and thus to preserve most of his usual driving task, at least at the operational
level considered here. Besides, simulated distraction task, defined as a
punctual appearance of a reading task, is quite close to real distractions
sources (navigation tasks, audio, inattention, etc.). Thus, in spite of the
simulation bias, the representativeness of the experimental situation allows to
access, to some extent, to the drivers subjective point of view.
Study of acceptability of driving assistance devices through the evocation of
lived experience was particularly rich. In particular, the several dimensions of
acceptability (Feeling/Interpretation/Utility/Attitude) underlined were useful to
compare modalities in details. For example, the speech relative to feeling and
interpretation were widely positive for sound, and negative for motor priming;
while the perception of utility was lightly positive for both sound and motor
priming.
Overall, assistance devices comparison shows the superiority in terms of
acceptability of auditory warning with respect to motor priming. On one hand,
ideal experimental conditions permitted to perceive the sound well and therefore
auditory warning was ranked at the head of preferences. Nevertheless,
participants had some doubts regarding “easy” perception of sound when
driving in a real car, and in real conditions; this may explain why auditory
warning appeared mainly in second choice (50 % in rank 2, fig.1.). On the other
hand, reproduced sound, which reminded of noise induced by driving on rumble
strips, strongly contributed to the acceptance of tested auditory warning. These
findings confirm the influence of sound design on acceptability.
Nevertheless, asymmetric steering wheel vibration (motor priming), which was
supposed to be more efficient was confirmed as it induced the best successful
correction. However, this assistance was not fully accepted by drivers, due to
317
Human Centred Design for Intelligent Transport Systems
the location of the stimulation. Indeed, most drivers believe that steering wheel
represents a prominent component to control vehicle and they were not very
confident in the use of this device in real driving conditions: in a real car,
vibrations are transmitted to steering wheel, and their amplitude and frequency
depend on road type, speed, vehicle, etc. This may induce unsafety feelings
and discomfort when using motor priming because drivers could not only
confuse whether vibrations are related to infrastructure or to assistance device,
but also lose control of their car.
Concerning the combined device, results in terms of effectiveness are
equivalent to those found with the motor priming device alone. In terms of
acceptability, some subjects declared that auditory warning helped them to
better interpret motor priming stimulation: “I do not know if it is the combination
of sound and steering wheel, but it seemed for me easier to manage”; “maybe
thanks to sound, I was able to distinguish the first priming vibration. I do not
know why it was easier for me to feel it”. Nevertheless, for others, nonsynchronization signals, bad lateralization of motor priming device and
confusion risks influenced negatively their user’s representation.
To conclude, the initial hypothesis on the combined mode is still reliable
because the association of a sound, accepted by the majority of drivers, with
efficient asymmetric vibrations on the steering wheel, seems to be a good
compromise to combine effectiveness and acceptability. Nevertheless, a better
integration of these assistance devices (by improvement of the synchronization
frequency between sound and haptic stimulation for example) is still necessary
to improve acceptability with preservation of device effectiveness. It should be
noted that this compromise is difficult to obtain.
Effectiveness, which has generally been considered as a fundamental criterion
for acceptability and usability in the Human-Machine Interaction domain 1 ,
appears not to be the most important factor for Human-Machine cooperation.
Indeed, driver did not immediately perceive the beneficial effect of the
assistance, and thus acceptability was not influenced. Besides, interferences
management, one of the important criteria of Human-Machine cooperation, can
induce annoyance even if performances are raised. However, this criterion of
effectiveness is essential in conception choices. This paradox makes important
to discuss where to place effectiveness in the user-centered conception loop. It
could be overcome by a compromise between device effectiveness
(performances measures for example) and acceptability criteria as defined in
this study.
5 CONCLUSION
Evaluation of acceptability in mutual control modes, through the analysis of
spontaneous and provoked verbalizations, evidenced which modality was
preferred and which one was rejected. In addition, these data permitted to
understand individual and collective reasons of these differences. None of all
assessed modalities was found to make unanimity of preferences, and drivers’
1
Because of the narrow link between perceived and real effectiveness.
318
Drivers’ acceptance of assistance functions
opinions were various. However, it clearly appeared that: (1) auditory warning
device was preferred although classified as second choice and (2) motor
priming device was massively rejected because of various reasons, such as the
location of the stimulation. Assistance devices’ effectiveness did not influence
drivers’ choices in terms of acceptability, and hence effectiveness could not
constitute the only criterion to choose cooperation mode.
In this paper, we underlined the potential gap between acceptability and
effectiveness. This point can be a crucial one for an efficient design of the
assistance. Consequently, we have to find the best compromise between these
criteria. In our study, the results of the combined device (sound and motor
priming) could be a possible solution for such compromise: the result of
effectiveness is good in terms of effectiveness and quite good in terms of
acceptability.
Furthermore the detailed analysis of verbalizations showed that for a majority of
subjects the combined device was not felt as a unified sensation but as two
distinct ones. This is the main explanation for the relative poor result of the
combined device in term of acceptability. Further studies should address this
major point by improving the accuracy of the synchronization of those
sensations.
6 Acknowledgements
Special Thanks to Béatrice Cahour for her precious advices concerning the
verbal data gathering.
7 References
[1]
[2]
[3]
[4]
[5]
[6]
[7]
Navarro, J., Mars, F., Hoc, J.M., Boisliveau, R., et Vienne, F.: Evaluation of
human-machine cooperation applied to lateral control in car driving. Proc.
of the 16th World Congress of the International Ergonomics Association
(IEA), Maastricht, Pays Bas: Elsevier. 2006, pp. 4957-4962
Hoc, J.M. and Blosseville, J.M.: Cooperation between drivers and in-car
automatic driving assistance. Proc.CSAPC, 2003, pp. 17-22
Suzuki, K. and Jansonn, H.: An analysis of driver’s steering behaviour
during auditory or haptic warnings for the designing of lane departure
warning system. Jap. Soc. Automotive Eng, 2003, Rev. 24. pp. 65-70
Hoc, J.M., Mars, F., Milleville-Pennel, I., Jolly, E., Netto, M., and
Blosseville, J.M.: Evaluation of Human-Machine Cooperation Modes in car
driving for safe lateral control in bends: function delegation and mutual
control modes. Le Travail Humain, 2006, (69), pp.115-185
Nielsen, J. Usability Engineering. Academic Press, Boston. 1993
Hoc, J.M.: Towards a cognitive approach to human-machine cooperation
in dynamic situations. International Journal of Human-Computer Studies,
2001. (54), pp. 509-540
Cahour, B.: L’expérience vécue des utilisateurs : pourquoi l’étudier et
comment ?, Actes du colloque CITE 2006 (Coopération, Innovation,
Technologie), Semaine de la connaissance, Nantes 26-30 juin.
319
Human Centred Design for Intelligent Transport Systems
[8]
[9]
[10]
[11]
[12]
[13]
[14]
Cahour, B.: Accès aux affects en situation professionnelle. Actes Epiques,
2003, pp. 108-110
Lefeuvre, R., Bordel, S., Guingouain, G., Pichot,N., Somat, A., and Teste,
B.: La mesure de l’acceptabilité sociale d’un produit technologique :
l’exemple des dispositifs d’aide à la conduite : Nouvelles technologies,
sécurité et exploitations routières. Revue générales des routes et des
aérodromes, 2004, N°832, pp.27-32
Stephan, K.L, Smith, S.E., Martin, R.L, Parker, P.A. and McAnally, K.I.:
Learning and retention of associations between auditory icons and
denotative referents: implications for design of auditory warnings. Human
Factors, Summer 2006, Vol 48 N°2, pp.288-299
Vermersch, P.: L’entretien d’explicitation. ESF ed., 1994
Vermersch, P., Martinez, C.M., Maurel, M., et Faingold, N.: Etude de l’effet
des relances en situation d’entretien. Revue Expliciter du GREX, (49), pp.
1-42
Navarro, J., Mars, F., Forzy, J.F., El-Jaafari, M., et Hoc, J.M.: Objective
and subjective assessment of warning and motor priming assistance
devices in car driving. Proceedings of the European Chapter of Human
factors and ergonomic society, Braunschweig, Germany. (in press)
Dillon, A., and Morris, M.: User Acceptance of new information technology:
theories and models. In M. Williams (ed.) Annual Review of Information
Science and Technology, Vol. 31, Medford NJ: Information Today, 1996,
pp. 3-32
320
Drivers’ acceptance of assistance functions
REFLEKTAS – USAGE OF REFLEXIVE DRIVER
REACTIONS IN ADAS-DEVELOPMENT
Ingmar Ehrenpfordt (DLR, Germany),
Anke Kullack (Braunschweig University of Technology, Germany),
Frank Eggert (Braunschweig University of Technology, Germany)
ingmar.ehrenpfordt@dlr.de
ABSTRACT: In common lane-keeping systems drivers are warned of lanedepartures by acoustic, visual or haptic devices. Thus, an association between
the signal and the desired action has to be established. Because the
association is vulnerable to interference and decay, the functionality of cognitive
systems is suboptimal. The new approach uses behavioural strategies. By
eliciting a steering-reflex, the attention of the driver is attracted and the correct
reaction is initiated. This principle was explored with 23 participants in a driving
simulator. First results already demonstrate the functionality of the principle.
Eliciting the appropriate reaction prevented road departures
1 Introduction
One goal which was set by the European Commission is to halve the number of
road fatalities between 2000 and 2010. They want to achieve this by
harmonisation of penalties and they promote new technologies to improve road
safety [1]. Other countries have their own, smarter goals like the “Vision Zero” in
Sweden [2]. One way to improve road safety is to develop new driver
assistance systems in order to prevent fatal crashes. Because lane departures
are a major cause of fatal crashes, many lane-keeping systems are developed.
2 The cognitive and the behavioural approach in the design
of Lane Keeping Assistance Systems
Classical cognitive systems use acoustic, visual or haptic devices to warn the
driver. Visual information of the vehicles status could be found everywhere in
the cockpit and centre console. Different blips are used for different acoustic
information or warnings. In case of the haptic devices, for example some car
manufacturers produce a vibration at the steering-wheel [3] or on one side of
the driver’s seat [4]. And there are some researches on assistants which use a
smooth force on the steering wheel in the direction of the middle of the lane.
In doing so an association between the signal and the desired action has to be
established in the context of learning processes. The drivers have to learn the
different associations. These also vary between the different car manufacturers.
The associations are subjected to interference and decay. Therefore the
functionality of these cognitive systems is suboptimal.
321
Human Centred Design for Intelligent Transport Systems
In contrast, our new approach uses behavioural strategies to prevent road
departures. Established, stable and irreversible stimulus-reaction links which
are involved in reflexes are used, which lead to a better reliability, even in
situations with a higher workload. There is no need for extra attention to
process the new incoming information by the driver.
Furthermore the reaction times are much shorter – reflexes have the shortest
latencies with 20-120ms – as reflexes involve other neural structures than
cognitive processes. There is no moment of shock which delayed a driver’s
reaction, because the reaction is reflexive.
As Figure 1 illustrates, reflexes comprise the brainstem and spinal cord,
whereas consolidated behaviour and intended actions involve higher regions
like the cerebellum or the association cortex. Consequently cognitive
information processing requires more time to produce an action in comparison
to reflexive reactions.
Fig.1. Scheme of the different levels of the sensomotor process and neural
structures, which are involved in these processes. The suggested hierarchical
order is simplified. In favour of a better illustration many afferent and efferent
connections are ignored [5]
3 The new approach of Lane Keeping Assistance
The aim of the development of the new lane keeping assistance is to overcome
the disadvantages of conventional cognitive systems. Primarily we want to
achieve faster reaction times and less incorrect reactions. The innovative idea is
to elicit a steering-reflex by a jerk at the steering wheel to initiate the correct
reaction and simultaneously enhance the driver’s attention. The drivers’
reflexive response starts even before he recognises what he is doing. But at this
322
Drivers’ acceptance of assistance functions
moment of recognition he is still turning the steering wheel in the right direction.
He has only to control the end of the reaction and fulfil the task.
The jerk is triangular and symmetric and proceeds into the same direction in
which the driver begins to leave the lane, thus a reflexive reaction against the
deviation is provoked. This principle was successfully analysed in several
studies in a driving simulator at the German Aerospace Center.
4 The Studies
The first of a number of studies explored systematically different adjustments of
the jerk to find the most effective one. Therefore a high number of variations
between the strength and the duration of the jerk have been tested. Afterwards
the jerk which was identified as the most appropriate was further examined. A
total of 23 participants took part in the study. The average age of the sample
was 40 years (SD=11). They drove for about one hour on a simulated rural
road.
Figure 2 shows the HMI-Lab at the German Aerospace Center which is used in
the studies.
Fig.2. The HMI-Lab at the German Aerospace Center
The usual cause for failures in lane keeping is distracted attention of the driver.
Thus, we simulated such a condition by presenting a distracting task to the
driver. While the drivers worked on this task, the system simulated a tendency
of the car to leave the lane on one side by means of adding a defined steering
angle. The distracting task was realized in two different forms, simulating
different forms of impaired attention.
One half of the participants had to track a ball which moved across the scene
with their eyes. The ball changed its size from time to time. The participants
323
Human Centred Design for Intelligent Transport Systems
were asked to indicate such an event by means of two buttons on the steering
wheel.
The other half of the participants had to solve easy mathematical tasks which
appeared on a screen in the centre console. These drivers had to indicate if a
presented solution was right or wrong with the same two buttons on the steering
wheel, Due to the distraction tasks, the participants did not recognise the
tendency of the car to leave the lane. As the car reached a pre-defined distance
to the lane-edge, the assistant was activated.
Afterwards the additional steering angle was deactivated in order to avoid any
artefact in the steering movements due to it. The subjects continued normal
driving and after a few minutes, the process started again.
5 Results
All studies which have been carried out demonstrated the functionality of the
principle. The jerk produced the reflexive response regularly and immediately
after the assistant has been activated. The participants reacted with a reflexive
steering-movement directed to the correct side. The compensatory corrective
actions following the reflex did not build up and the participants did not overcompensate. Moreover there was no steering in the wrong direction. This also
indicates that the jerk that was used was able to evoke the desired reflex.
Figure 3 shows an idealised reaction pattern for a departure to the right side of
the lane.
Fig.3. Idealised reaction pattern, explanation of the different measures
In order to better characterise the reaction, different measures were calculated.
A difficulty in the analysis of the reaction is that the participants react so fast
that the reset of the steering wheel produced by the system merges into the
start of the participant’s reaction. Therefore no exact reaction time can be
324
Drivers’ acceptance of assistance functions
calculated but only the maximum of the reaction time, which surely describes an
event quite a while after the reaction started. The maximum reaction time of
course is a very conservative measure and heavily overestimates the actual
reaction times.
Another important measure is the time from the onset of the assistant until the
car changes the direction. Furthermore the primary reaction (the reflex) and the
secondary reaction (compensatory reaction) have to be described. At the end of
the second component of the compensatory reaction which could be described
as an idealised damped oscillation the car follows its usual track in the middle
(or nearly in the middle) of the lane.
In addition to the objective data from the simulation, subjective data was
collected. After each study the participants were asked if they liked the system
and felt comfortable with it. In the first studies the participants stated that the
system helped them not to leave the roadway and that they felt safe with the
system.
Detailed analyses of the gathered data are under way.
6 A look into the future
Further studies will examine the reactions of drivers with different steering
modalities, like holding the steering wheel with only one hand.
Also the status of the drivers’ vigilance will be varied. Could the reflex triggered
with the accurate result during microsleeps?
The usefulness of a number of systems designed to apply principles of our
behavioural approach is currently under examination and will be presented in
the future.
7 References
[1]
[2]
[3]
[4]
[5]
European Commission: ‘White Paper’ (2001)
Swedish National Road Administration: ‘Vision Zero – from concept to
action’ (2006)
www.trw.com/images/cha_driver_assist_deutsch.pdf accessed 14.08.2006
www.citroen.com/CWW(en-US/TECHNOLOGIES/SECURITY/AFIL/
accessed 14.08.2006
Müsseler, J. and Prinz, W.: ‘Allgemeine Psychologie’ (Spektrum
Akademischer Verlag, 2002)
325
Drivers’ acceptance of assistance functions
MEET THE DRIVER NEEDS BY MATCHING
ASSISTANCE FUNCTIONS AND TASK DEMANDS
Astrid Kassner (DLR, Germany)
Astrid.kassner@dlr.de
ABSTRACT: Test drivers drove in fog as well as with clear visibility. They were
assisted with informing, warning or intervening Driver Assistance Systems or
drove without assistance. Driving data and acceptance judgments show the
best safety and acceptance values for the Info-Assistant. The Info-Assistant
allowed the consistent driving behavior, the safest headway parameters and the
shortest danger exposure. Acceptance evaluation showed that the InfoAssistant provided the most relief and maintained control over driving at the
same time. Paying attention to task demands gives some useful information
about the user needs. Matching of task difficulty and assistance functionality
seems to be a useful step within the development of Driver Assistance
Systems.
1 Introduction
Advanced Driver Assistance Systems (ADAS) are supposed to increase traffic
safety. Although a lot of systems are available the take rate with 1% of all cars
is low [1]. Therefore positive consequences for traffic safety are small and
should be enhanced. Maybe car drivers don’t know the advantages or the
functionality of these systems. At the same time with the introduction of active
systems (such as lane departure warning) there will be more and more systems
that involve the driver and need his understanding of their functioning. This
requires more precise knowledge about the cooperation of driver, car and
ADAS than with passive systems (ABS e.g.). The development of ADAS
motivated by economic or technical aspects doesn’t seem to be very useful.
Instead, the development process should focus on the driver and his needs to
get some assistance. This assistance needs arise mainly from the difficulty of
the driving task since task difficulty is a main determinant of driver behavior [2].
According to the Information Processing Approach (e.g. [3]) the difficulty can
consist in perceiving the relevant information, deciding or executing the action
correctly. ADAS could then assist through information, warning or intervention.
For example, if perception in fog is difficult an ADAS could inform you about
relevant things in your near environment, could warn you and facilitate your
decision or could brake by itself. It would be useful to know if matching of the
functionality of ADAS (informing, warning, intervening) with task difficulty
(perception, decision, action execution) results in better driving performance
and system acceptance. Traffic safety could profit because systems would be
more efficient and then maybe more systems would be used because of the
better acceptance. Therefore this knowledge could be useful for the
development process of future ADAS.
327
Human Centred Design for Intelligent Transport Systems
2 Method
We examined this question with the Dynamic Driving Simulator of the DLR. 16
test persons participated aged between 24 and 39 years (mean = 30.1 years ,
STD = 4.85 years). Participation was voluntarily and compensated. Before the
experiment each driver completed a training session in the simulator. In the
experiment the subjects had to drive on a straight road for 10 minutes and
repeatedly approached a lead car which drove with a constant speed of 50
km/h. After at least 20 seconds each lead car pulled off the road and the
participants had the chance to drive unaffected until they approached the next
lead car. The distances between the cars within one trial varied between 900
und 1600m. The order of the distances varied from trial to trial to inhibit that the
drivers build some expectations about the appearance of the next car.
As independent variables we varied the task difficulty and the functionality of
assistance. Task difficulty was varied by creating two different perceptional
conditions. In the easy task condition the drivers had clear visibility conditions.
In the difficult task they drove in fog. In both conditions they were instructed to
drive safely and as fast as possible. To prevent any velocity reduction especially
in the difficult condition which might reduce task difficulty we additionally
motivated the test persons to drive fast by time measurements. The assistance
functionality was varied by supporting the drivers either by an informing,
warning or braking Assistant or let them drive without assistance. The
experiment was designed as a within-subjects design. Thus each driver drove in
clear visibility with either an informing, warning or braking Assistant or without
assistance and each driver drove in fog with either an informing, warning or
braking Assistant or without assistance.
To let the test drivers familiarize themselves with the Assistants and to minimize
any carry-over effects each trial was performed twice and just the second trial
was used for the analysis. Half of the participants started with the easy task, the
other half started with the difficult task. The order of the assistance conditions
within the difficulty condition was counterbalanced.
The different Assistants were supposed to support different parts of the driving
task. The Info-Assistant is supposed to support the perception of the headway
to a lead car since it always shows a topview of the road and a possible lead
vehicle in a head-up display. The Warning-Assistant gives an auditive and
visual signal depending on the collision risk (based on the Time-To-Collision)
and is supposed to facilitate the decision to brake. ). The Brake-Assistant
brakes if the collision risk is imminent to prevent any collision and gives some
visual and auditive feedback. Figure 1 shows the three Assistants.
328
Drivers’ acceptance of assistance functions
Fig.1. Info-Assistant, visual feedback of the Warning-Assistant and visual
feedback of the Brake-Assistant, trial with clear visibility
The three Assistants were activated at different times. The design emerged
from preliminary expert tests. The Info-Assistant was always present and
started to display the lead car very early to give the driver enough time for the
next steps to take. The Warning-Assistant was active when a certain threshold
for the time-to-collision was reached. It warned early enough to allow the driver
a moderate braking action but it warned late enough not to anticipate the
driver’s decision to brake. The Brake-Assistant intervened as the latest when
the time-to-collision was small. It braked early enough to avoid a collision but
late enough to allow the driver braking by himself.
3 Results
Three of the ten approaching events of each trial were chosen because then the
distance to the next lead car was large enough and there were no missing data.
For each of this three events the driving behavior before the approaching
(phase 1) and during the approaching process (phase 2) was examined. Phase
1 lasted 7 seconds and ended when the lead car was 6 seconds (time
headway) away. Phase 2 lasted at least 20 seconds depending on the driving
behavior and ended when the lead car left the road to the right. The activities of
the Assistance Systems started during phase 2, the always present InfoAssistant just showed an empty road in phase 1. The results were analyzed
using a Two-Way Analysis of Variance with repeated Measurements with the
factors task difficulty (clear visibility and fog) and Assistance System (without
assistance, Info-Assistant, Warning-Assistant and Brake-Assistant).
First I will focus on phase 1 and start with the speed behavior. As desired there
were no velocity differences between difficulty conditions and Assistance
Systems in phase 1. Thus the drivers didn’t reduce the task difficulty of the fog
condition by reducing the velocity. Also no Assistance System induced a lower
or higher velocity either. A meaningful variable of the speed behavior is the
speed variability. It describes how much the drivers changed their speed and
showed a main effect of the task difficulty (F(1,14) = 14.97, p< .05). With clear
visibility velocity changed more. Mean deceleration showed the same effect
which means that there was more braking activity under clear visibility
conditions. A reason might be that the lead car in the clear visibility condition
was already visible in this phase and the drivers started decelerating.
Furthermore speed variability showed a main effect of the Assistance System
(F(3,42) = 5.37, p< .05) with the Info-Assistant leading to the most constant
speed behavior. Again, mean deceleration was responsible which was the
329
Human Centred Design for Intelligent Transport Systems
lowest with the Info-Assistant. Driver behavior in phase 1 is also useful to
examine possible behavioral adaptations since drivers in this phase drove
unaffected by any lead car. Therefore we compared the mean velocity of the
first event with the mean velocity of the third event and found no differences
(F(1,13)= 0.396, p= 0.54). Thus there were no short term behavioral
adaptations.
Phase 2 showed no velocity differences between the visibility conditions or the
Assistance Systems as well. As instructed the test persons drove with similar
velocity in clear visibility as well as in fog. This time speed variability in fog was
higher than with clear visibility conditions (F(1,14) = 8.92, p< .05). This again
resulted from the deceleration behavior which showed the same main effect. In
phase 2 braking in fog was more severe than in clear visibility and the drivers
decelerated to a lower velocity. The speed variability depended also on the
Assistance System (F(3,42) = 3.06, p< .05). The Info-Assistant allowed the
smoothest speed behavior. Especially in fog drivers reduced their speed with
the Info-Assistant not to the same extent as with the other Assistants. Figure 2
shows speed variability as a measure of driving behavior.
Fig.2. Speed variability in phase 1 and 2 [m/s]
This speed behavior can result in safe or unsafe traffic situations that are
especially in phase 2 a matter of interest. An important safety indicator is the
time-to-collision (TTC) that contains the time until a possible collision occurs
and measures therefore how dangerous a situation is. Traffic situations with
less than 4 seconds TTC are considered to be dangerous. The proportion of the
dangerous time (TTC) to the time of the whole phase describes how long
danger existed. This time-exposed TTC shows a main effect of task difficulty
(F(1,14) = 26.94, p< .001), which means that the danger in fog existed for a
longer period of time than with clear visibility. Other safety indicators showed
the same effect. Distance headway or time headway describe how closely the
driver followed the lead car. In fog the mean time headway and the mean
distance headway were smaller than with clear visibility. Furthermore the timeexposed TTC depended on the interaction of visibility condition and Assistance
System (F(3,42) = 5.34, p< .05). With the Info-Assistant in fog drivers were not
as long exposed to the danger than with the other Assistance Systems.
330
Drivers’ acceptance of assistance functions
Distance headway and time headway were the safest for the Info-Assistant as
well. However, the duration of lane exceedences was the longest with the InfoAssistant (F(1.5, 21.1) = 4.95, p< .05) but happened mainly with clear visibility.
Figure 3 shows time-exposed TTC and lane exceedences as measures of traffic
safety.
Fig.3. Time-exposed TTC and duration of lane exceedences
Acceptance was measured with questionnaires after each trial. Driver judged
the speedometer in the trial without assistance instead of an Assistant System.
With judging the content of the Assistant drivers evaluated how useful
information about the distance to the lead car is compared to a collision warning
or a collision-dependent intervention. Results showed an interaction of task
difficulty and Assistance System (F(3,42)= 4.67, p< .05). With clear visibility
information about the distance made no sense, but in fog it’s usefulness was
similar to a warning or intervention. Evaluations of the user interface tended to
depend on the interaction of both factors (F(3,42) = 2.61, p= .064). The user
interface of the Info-Assistant performed poorly with clear visibility but as good
as the other Assistants in fog. The relief through the use of assistance showed
an interaction, too (F(3,42) = 2.9, p< .05). The test drivers evaluated whether
distance keeping to the lead vehicle was easier or more difficult when using the
Assistant. The Info-Assistant made this task easier, especially in fog. Control
over driving is another important variable and describes how the use of an
Assistant changes the own control over driving. It was evaluated dependent on
both factors (F(3,42) = 4.85, p< .05), which means that the Info-Assistant in fog
reduced the control least of all. However, the distraction through assistance
showed a main effect (F(1.7,23.8) = 9.94, p< .001) with the Info-Assistant
distracting most. Figure 4 shows the important evaluations.
331
Human Centred Design for Intelligent Transport Systems
Fig.4. Evaluation of relief, control over driving and distraction. A score of 15
means the best evaluation
4 Discussion
In the experiment the participants drove on a straight road with good or bad
visibility conditions and therefore different task difficulties. They were supported
by one of three Assistance Systems that differed in the kind of functionality or
they drove without assistance. Following our instruction they drove in both task
difficulty conditions with the same velocity. Thus the drivers didn’t reduce task
difficulty of the fog condition and they experienced a more difficult driving task in
the fog condition than with clear visibility. Task difficulty further resulted in
different driving behavior. In phase 1 deceleration was stronger with clear
visibility since they already saw the lead car, whereas in phase 2 they had to
decelerate stronger in fog. The difficult fog condition resulted also in less safe
traffic situations. Drivers then had lower safety margins and dangerous time
periods were longer. More relevant than task difficulty effects are the effects of
the different Assistance Systems. No Assistance System tempted the drivers to
drive faster. The test persons drove more consistent with the Info-Assistant than
with the other Assistants since they didn’t have to decelerate so much. In phase
2 the Info-Assistant especially in fog generated less deceleration. Thus the InfoAssistant allowed a very smooth driving behavior with early and weak
decelerations. The Info-Assistant also resulted in the best traffic safety values
especially in fog. Dangerous time periods with the Info-Assistant were shorter.
Distance and time headway were safer. Acceptance evaluation in the fog
condition didn’t differ between the Assistance Systems regarding the delivered
content (distance information, collision warning or collision-dependent
intervention) and the user interface. The major subjective advantages of the
Info-Assistant consisted in the usefulness. The Info-Assistant allowed the most
control over driving which corresponds with the better anticipative abilities in
driving behavior. In addition the Info-Assistant facilitated distance keeping most
which corresponds with higher traffic safety in driving behavior. However,
drivers evaluated the Info-Assistant as most distractive which is supported by
the time of lane exceedences. Since the Info-Assistant doesn’t result in more
variation in the lateral position than the other Assistance Systems, the
distractive effect should be of more subjective than objective nature. Overall the
332
Drivers’ acceptance of assistance functions
Info-Assistant shows the best driving behavior, the highest traffic safety and the
best acceptance evaluations.
Our study shows that the longitudinal control in a perceptually difficult driving
task is best assisted by an Assistant that provides early information about a
lead car. The Warning-Assistant which gave a warning about some danger with
the request to brake or the Brake-Assistant which intervened in case of an
emergency resulted in less safe traffic situations and these Assistants were also
less accepted. This happened despite the fact that these systems also gave the
driver enough time to brake, thus didn’t induce any time pressure. Apparently it
is most useful for the driver in this situation to get some information about future
events which allows the driver to anticipate his behavior. With the different
Assistance Systems we also tried to vary different Levels of Automation. The
Info-Assistant represents the lowest level since it just gives some relevant
information about the environment and the driver has to deal with the
consequences. The Brake-Assistant represents the highest Level of Automation
since it acts for the driver. The study shows that the highest degree of
automation is not always desirable. One has to consider also the task difficulty.
On the other hand we do not know if the drivers really used the Assistance
Systems as they were intended to use. For example it is quite possible to use
the Brake-Assistant as a warning when the start of the system’s activity is used
as a hint to start with braking. Similarly it is possible to use the Info-Assistant as
a warning when one specific status of the display is used to start braking. We
hope to get some insights into this issue when we evaluate the interviews where
we asked the participants how they used the systems. Another weakness of the
study is that the functionality of the Assistance Systems is confounded with the
kind and number of the used sensory channels. The Info-Assistant just uses the
visual channel, the Warning-Assistant additionally uses the auditory channel.
The Brake-Assistant uses the visual, auditory and the haptic channel. However,
the used channel results from the functionality of the assistance. One cannot
combine each functionality with each channel, e.g. an intervention must use the
haptic channel and can’t be visual. This restriction means that we can’t be sure
that the advantages of the Info-Assistant only result from it’s informing
character, maybe the used channel is also responsible. However, we found no
differences in the acceptance evaluations regarding the used channels. As
another possible weakness the functionality of the Assistance Systems is
confounded with the different times they started to be active. Again this results
from their functionality. The Info-Assistant has to be active early to allow the
driver all the other steps to take. A warning as early as this information would
have been considered as a false alarm since the lead car was far ahead. A
braking at this early time would be completely senseless. There exists no
common point in time to start the activity of the different functionalities. The
functionality cannot be separated from the starting point. In other words – the
starting point is in addition to the functionality part of the Assistance System.
One might argue that the smoother and safer driving behavior of the InfoAssistant results from it’s earlier activity. Of course it allowed an earlier reaction
of the driver but this earlier reaction wouldn’t have happened if the drivers
hadn’t thought that an anticipative behavior is useful. Finally we have to
consider that the identified effects of the Assistance Systems may change with
longer use. We found no short-term adaptations in terms of velocity changes
333
Human Centred Design for Intelligent Transport Systems
after the participants had used each system at least 20 times. Longer term
adaptations can usually be measured only after a familiarization period.
Possible determinants of behavioral adaptations are the influence of the system
on the way the driving task is performed, the drivers’ possibilities to change their
behavior and the presence of competitive motives for changing behavior [4].
Maybe it can be hypothesized that a lower Level of Automation changes the
way a task is performed to a less extent and therefore induces less behavioral
adaptations than higher Levels of Automation. But this has to be examined in a
long-term study.
5 Acknowledgements
This research was supported by the Lower Saxony Ministry for Economics,
Labour and Transport within the project “Strategies for Driver Assistance
Systems”.
6 References
[1]
[2]
[3]
[4]
Mercer-Untersuchung, Automobile Sicherheitstechnik. Available under:
http://www.presseportal.de/story.htx?nr=553838&action=bigpic&att=28898,
accessed May 2006
Fuller, R. (2005). Towards a general theory of driver behaviour. Accident
Analysis and Prevention, 37, pp. 461-472
Timpe, K.-P., Giesa, H.-G. & Seifert, K.: Engineering Psychology. in: C.D.
Spielberger (Ed.): Encyclopedia of Applied Psychology (Elsevier, 2004,
Volume I), pp. 777-786
Saad, F. et al. (2004). Literature review of behavioural effects. EU project
AIDE, project IST-1-507674-IP, Deliverable 1-2-1
334
Drivers’ acceptance of assistance functions
PERCEPTIONS OF RISK AND CONTROL
UNDERSTANDING ACCEPTANCE OF ADVANCED
DRIVER ASSISTANCE SYSTEMS
Somya Joshi (ICCS, Greece), Angelos Amditis (ICCS, Greece), Thierry Bellet
(INRETS, France), Vanessa Bodard (INRETS, France)
s.joshi@iccs.gr
ABSTRACT: This paper reports on a part of a larger study carried out within the
HUMANIST network for excellence, within the remit of a post-doctoral research
project. It examines the relationship between driver perceptions of risk and
control, and the influence this has on their acceptance of advanced driver
assistance systems. The paper describes in depth the unique methodology
employed, the experimental design and the analytical framework used. It builds
on current knowledge in the area of ADAS acceptance and behavioural
adaptation resulting from it.
1 Introduction
With a marked increase in advanced driver assistance systems (ADAS) being
designed and deployed for cars, there is a logical emergence of studies that
critically examine the influence these have on driver behavior and attitudes
towards risk and safety. The research question addressed within this paper
asks to what extent the level of perceived criticality or risk on the part of drivers
influences their acceptance of advanced assistance. The hypothesis thus tested
the relation between increased perception of personal control in a given
situation and the influence this has over the acceptance of risk by individuals,
as well as on their subsequent acceptance of ADAS. In relation to driving
behaviour, by risk perception here we refer to “the subjective experience of risk
in potential traffic hazards” [1]. Individual road users experience or anticipate, at
any moment in time, a certain amount of loss of control, and they compare this
with their target level of risk.
Different research traditions have attempted to explain individual differences in
risky driving behaviour and accident involvement. Presently, the technological
feasibility of most ADAS is not the main issue for implementation anymore [2].
In fact, the first ADAS applications have already entered the market, such as
adaptive cruise controls and collision warning systems. The focus in scientific
research on ADAS in the past years has shifted from basic technology research
and development towards the complexity and impacts of implementation of
ADAS [3]. The present study contributes to this by undertaking an analysis of
driver identifications, motivations and preferences with regard to risk-taking
behaviour, and subsequently their acceptance of ADAS. The analytical
framework of Activity Theory was employed to systematically investigate the
complex relationships and tool-mediation that emerge within the present-day
augmented environment of intelligent transport. By focusing on the tools (both
335
Human Centred Design for Intelligent Transport Systems
technological and conceptual) that mediate between our subject group of
drivers and our augmented driving environment, this paper attempts to critically
examine how diverse driver attitudes towards risk and control can be factored
into the design of intelligent in-car systems.
2 Methodology
This paper reports on the findings from a two-year research collaboration within
the HUMANIST network 1 The research took place over a period of a year and
half, with the support of infrastructure sharing in terms of both equipment and
knowledge expertise. An extensive literature review was conducted on the area
of risk perception, compensation and ADAS assistance. The development of
qualitative interview tools and questionnaires followed this. The uniqueness of
this research project lies in its methodology, which combines qualitative,
interpretative analysis tools, thereby allowing for a deeper, richer understanding
of driver decision-making behaviour and subjective attitudes towards risk and
safety.
2.1
Participants
A total of 20 subjects participated in our study, who were selected from a
diverse background, cutting across gender, age, driving experience, and license
history. With regard to the novice-elderly distribution, the subjects covered ages
ranging from 24 to 66 (see fig.1), their experience ranged from 1 to 48 years
(see fig. 2). In terms of mileage the subjects varied from below 3000 kms to
100000 kms. Finally, we had a mixture within the group of subjects that had
points on their license and those that held a clean license history.
Fig.1. Age/Gender distribution
2.2
Fig.2. Experience distribution
Simulator
The simulator we used was a static base one, with a vehicle interface driven by
a microcontroller, steering wheel, brake and accelerator with driven torque and
effort. The computers used for the simulation were 4 PC:
1
Between ICCS, Greece and INRETS, France {April 2006 – April 2008}
336
Drivers’ acceptance of assistance functions
•
1 PC for dynamic models of the vehicle and the traffic: PC 300Mhz
•
3 PC with high performance graphic card for the visuals (front views and
rear views of the road scene): PC 300 Mhz
The front field of view for the Bron simulator was 150° horizontal (on 3 screens)
and 40° vertical. The rear-view was embedded in the front view.
2.3
Experimental Design
The simulator part of the experiment was divided into three main stages. These
were: Orientation, Non-assistance and Assistance. As the name suggests,
during the first stage, we oriented the subjects with the two road environments
(i.e. urban/motorway) and allowed them to familiarise themselves with the car
and the simulator itself. This was followed by two series (1&2) where the driver
experienced a range of critical and non-critical situations, but without any
automation or assistance from the intelligent vehicle. Then in series (3&4) they
once again experienced a range of critical driving condition, with assistance in
the form of automatic breaking, steering control and speed reduction. Warning
assistance was given by way of audio (beeps) and visual (flashing diode)
signals. Finally the last section of series 4 ended with a near-collision scenario.
In addition to variation in terms of degree of criticality, type of assistance, and
road environment, we also included dynamic agents within the simulation such
as pedestrians crossing the street and motorcyclists. The order of the sections
was rotated at random between subjects to counterbalance possible learning
effects. At the end of each scenario there was a freeze – where the subjects
could assess the situation and their response. We experienced no cases of
simulator sickness or dropping-out within our sample group.
We designed a questionnaire tailor made for investigating abstract attitudes and
subjective perceptions regarding risk and control. The questions varied in
response-format from those that required graded answers on a pre-determined
scale to those that required elaboration. We asked the subjects to fill out the
first set of questions regarding self-assessment before they entered the
simulation scenarios, so as to capture their perception prior to engagement with
ADAS. Then during the scenarios we asked questions specific to situation
awareness and usability during the freezes. After the simulator part, we again
asked our subjects to fill in questionnaires and participate in semi-structured
subjective interviews, where they had another opportunity to provide rich data
on their perceptions of risk and control and their subsequent acceptance or
need for ADAS. Apart from Automatic Breaking Systems, none of the subjects
were familiar with the advanced assistances systems that we were testing.
The interview part of the experiment was divided into two main stages, namely
pre-simulation and post-simulation. In the first stage, we interviewed the
subjects in depth regarding their self-assessment of personal driving skills as
well as identifications with driving styles. We further questioned them regarding
their attitudes towards risk-taking and control within the context of automation.
In the post-simulator stage, we interviewed the subjects in relation to their
perceived performance in the various situations; their awareness of risk, as well
as their acceptance of perceived assistance in those same situations. Their
interview feedback was recorded both on audio and video. One major limitation
337
Human Centred Design for Intelligent Transport Systems
of the experiment was its reliance on simulator tests alone and no trials on testtracks or real roads, especially given the objective of the study being to examine
perceptions of ‘risk’. However given the ethical constraints of placing subjects in
situations of high criticality, the simulator option emerged as the only feasible
one for the moment. As these systems are developed further and introduced in
the market, one can carry out field trials, in real traffic conditions.
3 Theoretical framework
Within the broader Human-Computer-Interaction community it has become
tacitly accepted knowledge that people cannot articulate what they are doing (a
notion sometimes used as a justification for observational studies and
sometimes used to avoid talking to users at all). Activity theory within this
context has something interesting to tell us about the value of interview data.
We have used this framework to critically analyse the complex and rich
information we’ve gathered from users about their perceptions and motivations
for using ADAS. The fundamental concept of the Activity approach was
formulated by Lev Vygotsky [4], who spoke of artefact-mediated and object
(motivation)-oriented action. At the apex of this model (see fig.3) lies the
mediating tools (instruments) which are situated between the group undertaking
the activity (subject) and their desired goals and motivations (object). While the
constraints and access points (rules) determine the interactions between the
subject group and the stakeholders (community), the hierarchies of power and
expertise within an organization (division of labour) mediate between the
stakeholder communities and the overarching objectives and outcomes of the
activity. It is important to note that an activity system is never static. Tasks are
reassigned and re-evaluated, rules are bent and reinterpreted. There is
constant movement between the nodes of the activity system. What initially
appears as an object may soon be transformed into an outcome, subsequently
turned into an instrument, and perhaps later into a rule [5]
Fig.3. The activity model (Engeström 1987, p.78)
“Tools” mediate the relationship between subject and object, while the
relationship between subject and community is mediated by "rules" and the
338
Drivers’ acceptance of assistance functions
relationship between object and community is mediated by the "division of
labour". What is of specific relevance to our study here is the way this model
enables us to examine the mediatory role played by ADAS and the influence
this has on driver behaviour.
4 Analysis
The environment that we refer to here concerns the physical features of the
road (weather, geometry, signs and signals), the driver’s own speed and
direction, and the paths and speeds of other road users. As subject here we
refer to the individual driver, while the instruments in question would be the
ADAS and IVIS functions available within the experimental car. Our explicit goal
here, or the object, would be to reduce accidents and injury on the road, thereby
making the overall environment safer by endowing the subject with more
informed decision-making powers. In Fig. 4 below we see this represented
within the framework of the activity model.
Fig.4. ADAS Activity model
The outcome of the activity however is determined by the interactions between
the various nodes. And given the subjective nature of risk it is not surprising that
the final outcome of the activity could take form either in line with the desired
object of activity or in tangent to it. For instance the perceived level of risk will
be relatively low if the driver is confident about having the necessary coping
skills, and higher in the case of those who doubt their abilities. This was
precisely what was reported by one of our subjects during his self-assessment
exercise:
“I consider myself a risk taker, however it is very important for me to be in
control. Being in control for me means being aware of what is happening around
me, to be at a speed that I can master and in general be in charge of the
situation.”
Thus our subject was implying that risk taking was acceptable, in so far as the
some of the variables were under his control. Taking this a step further, it is
logical to argue that individuals differ not only in the accident risk they are willing
to accept but also in their ability to perceive accident risk and in their decisionmaking and executive skills in the face of risk. In other words, people differ in
both willingness (i.e. acceptance) and ability (skill). Perceptual skill includes the
339
Human Centred Design for Intelligent Transport Systems
ability to correctly assess one's level of decision-making and vehicle-handling
skill. This is important, because it implies that persons with limited decisionmaking or vehicle-handling skills are at no greater accident risk, provided they
realize their limitations and act accordingly.
However as situation awareness varies amongst drivers, so does their subject
evaluation of the posed risk. Burger et al. [6], have found that those with a high
desire for control exhibited a greater illusion of control (perceived control over
chance events). In terms of Risk Compensation, this could mean the allowance
of additional risk accepted in line with increased perceptions of control. This is
what we referred to earlier as the tangential outcome that the activity could take,
where the subject (driver) interacts with his/her environment via certain tools (in
this case ADAS) that result in his/her adapted behavior compromising safety.
The theory of risk compensation suggests that safety measures which reduce
risk to levels below the setting of the “risk thermostat” will be countered by
behaviour which reasserts the levels with which people were originally content.
If the propensity, or willingness, to take risks is the principal determinant of the
accident rate then this rate can only be reduced by measures that reduce the
propensity.
The primary functionality of ADAS, as is understood at present, is to facilitate
the task performance of drivers by providing real-time advice, instruction and
warnings. This type of systems is usually also described by the term “co-driver
systems” or “driver support systems”. Driver support systems may operate in
advisory, semi-automatic or automatic mode [7], all of which may have different
consequences for the driving task, and with that on traffic safety. Although the
articulated object or goal of a driver support system is to have a positive effect
on traffic safety, unintended effects have been shown on driver behaviour,
indicative of negative effects on traffic safety [8]. Firstly, the provision of
information potentially leads to a situation where the driver's attention is
diverted from traffic. Secondly, taking over (part of) the driving task by a codriver system may well produce behavioural adaptation. This behavioural
adaptation, or compensation as it is called in a wider field, must be taken into
account when investigating the conditions for introduction of ADAS [9]. When
interviewed post simulation, one of our subjects outlined for us this very feature
of compensation.
“When a system adds something that I don’t have, for instance in the case of
fog, or night-time if a systems takes control, due to my inability to see well in
poor conditions, I can accept that.”
Would this mean that when normally the driver would drive cautiously in such
impaired conditions, he would now compensate and thus pay less attention to
his immediate environment due to the new options afforded to him via such
tools? The critical issue here is one of dependency on a technical artefact that
could potentially lead to overlooking crucial variables and affecting the
stakeholders in an adverse way. For instance there is now substantial evidence
that the effect of risk compensation has been to shift part of the burden of risk
from people in vehicles to vulnerable road users outside vehicles, leaving the
total number killed in road accidents that could be attributed to seat belt
legislation little changed [10].
340
Drivers’ acceptance of assistance functions
Real life situations always involve an intertwined and connected web of
activities, which can be distinguished according to their objects. Participation in
connected activities having very different objects can cause tensions and
distortions, as is seen in this case to be risk compensation. For instance the
intention or motivation of the ADAS designer would be to enable the driver to
better cope with impaired conditions such as fog, low visibility etc. However the
motivations underpinning the driver’s behaviour might be centered around
reducing mental load, and hence once the driver assumes that the ADAS will
automatically take over certain functions, he might alter or adjust his level of risk
acceptance thereby not achieving the intended object of enhanced safety.
Interestingly though perhaps not surprisingly the tolerance for errors in such
systems emerged as very low, as was articulated by our subjects:
“As long as the automation was doing something that had not occurred to me
before, I’d be glad that it was there. I am more grateful of automation of course
in critical conditions. But if it made a mistake even once, I would not use it, as
its very important that the car responds in a way that I want it to. A high level of
accuracy is imperative.”
In addition to the issue of accuracy, two other crucial points are brought forward
in the above statement. The first is the co-relation between perceived criticality
and the acceptance of assistance, and the second being the co-relation
between what action is performed by the ADAS tool and what is perceived as
‘want’ or motivation behind the subject or driver’s action. The acceptance of
ADAS was found to vary with driver control [11]; taking the driver out of the
loop was considered a problem by many (potential) end-users. An international
questionnaire survey carried out by Bekiaris et al, [11], indicated that the driver
population was reluctant to release vehicle control, but was willing to accept it in
emergency situations. We tested this on our sample group and received
responses in support. We start with an elderly driver in our sample who said:
“I am accustomed to being master of my car, so I would be very upset to let go
of decisions and have no control. However if I have no chance or options in
really critical situations, as in case for example of a big lorry losing control
behind me, I’d be happy with automation. Also when I haven’t seen a vulnerable
road user such as a motorcyclist or pedestrian, I would appreciate some
assistance.”
Again we see that the subject has accepted a given level of loss of control in
situations where the choice is limited and a lot at stake. There is a demarcation
made for acceptance in situations where there is clearly and explicitly a value
added by the ADAS tool. This sentiment was further reinforced by another
member of our sample who said:
“If the situation is under my control, I don’t appreciate the car trying to take a
decision. It disturbs me. But when I am a prisoner of a situation, i.e. I’m out of
options, then its ok by me.”
More specifically, regarding the issue of perceived criticality a study found that
taking over of control in case of short headway to a lead car was less
appreciated than warnings or suggestions of the appropriate action in a test of
different types of Collision Avoidance Systems [12]. Although drivers expect a
positive safety effect by this type of anti-collision systems and other forms of
341
Human Centred Design for Intelligent Transport Systems
ADAS, they have at the same time reservations against it. Handing over control
to a device and the automated braking function were evaluated as negative
aspects of ADAS systems [13]. In our sample this distinction was also
reinforced. For instance drivers clearly indicated which automation functions
they regarded as acceptable and which they were strongly against having in
their cars. What we found during the course of our experiment is that drivers
sought more a ‘confirmation’ of their assessment of criticality and subsequent
decision-making, than an outright decision from the system. A subject from our
sample neatly captured this widely held sentiment when he said:
“The efficiency of the assistance depends on how it helps you with a decision in
a critical time. So as in this case, I saw the light (diode warning) before I heard
the beep, it didn’t alarm me, but only confirmed for me that it was a dangerous
situation and the right decision to break.”
Thus by confirming an already held belief, the function truly assisted the driver
in question, rather than further expend his limited attention in a crucial moment.
The timing, sequence and intensity of this intrusion was critical in assisting him,
and thus not surprisingly was rated high in acceptability value, by him. While
more accepting of wider implementation of automation functions, another
subject specified the limitations brought by the environment:
“I do not mind if the automation takes control of the car in a motorway, but for
sure not in an urban environment, there are just too many variables. If an
automation does something unexpected, this can be a limitation, but again this
can be solved by training drivers and through education programs.”
In a recent study by Comte et al [14], it is demonstrated that drivers are
sensitive to the prevailing traffic conditions (environment) when deciding
whether to use an advanced assistance system. She investigated driver
personality type and the influence this had on their propensity to use such a
safety system. Towards this end she employed a questionnaire [2] that allows
drivers to rate the acceptability of various driver support systems. The results
from this on-road study indicated that those drivers who enjoyed and engaged
in speeding, or exceeding the speed limit (i.e. take risks), were less likely to use
a system that would inhibit this. This is an incredibly important finding when
considering the mechanisms for implementing ADAS: as those drivers who
would benefit most would be less likely to use a voluntary system. In order to
arrive at a better understanding of what makes drivers more accepting and
adopting of ADAS systems, we asked our sample group to qualify their choices
in terms of all they experienced during the experiment (automatic breaking,
steering control and intelligent speed reduction). One subject responded:
“Relevance of the assistance depends on a situation. Right now as my speed
was very slow, I felt in control, even though the situation was very critical. So, I
would definitely not want an automated function kicking-in at this point.”
Thus our subject qualified three things here. One, her awareness of the
situation; two, her subjective evaluation of its criticality, and finally three, her
decision to reject automation on grounds of her being in control. These complex
and often sub-conscious calculations occur in a dynamic manner within a
fraction of a second, thus not surprisingly our sample group had concerns about
the ability of an ADAS to respond in similarly complex ways. This brings us to a
342
Drivers’ acceptance of assistance functions
very important aspect in user acceptance of assistance, namely: Trust. This is
linked closely with the previously mentioned factor of accuracy or low tolerance
for errors in ADAS enabled vehicles. A member or our sample group brought
this to our attention when she said:
“In the case of automation, its very important for me to know that the system will
not do something opposite in reaction to what I would have done. It’s very
important to have this trust. So I would like a test period where the system
needs to prove to me that it can make the best choice and I will need to verify
this. The system will show me that it can balance pressure and speed in order
to do a much better job than me in a surprising (risky) situation.”
This emerged as a common thread resonating in the responses of a majority of
our sample group. The need for a match (i.e. its predictability of action) between
the driver’s instinctive response in a critical driving situation and the automation
response was identified as a major criterion for acceptance.
In figure 5 below we see the shifting nodes of activity with the ADAS context, as
represented in the subject responses we have analysed so far. To begin with
the subject group can emerge as the designers of ADAS who hold certain
assumptions regarding the object of their activity. This could be increased road
safety as we looked at before, or simply an acceptance and user market for the
systems they create. Also we see that the mediating factor has the potential to
shift from the ADAS tool itself to the trust that users can invest in the system,
and its predictability of response (in other words the way it matches the users
intuitive response). An increased reliance on such a system could potentially
lead to greater acceptance of risk and this could in turn have unintended effects
on the driver and other road user stakeholder group. Finally, even if the driver
group decide to relinquish decision-making control onto the system and accept
the system, there is no guarantee that a direct co-relation can be drawn
between the use of the system and any decline in accidents.
Fig.5. Dynamic nodes of activity
343
Human Centred Design for Intelligent Transport Systems
5 Conclusion
To sum, what we have covered here is a rich and complex terrain of driver risk
perception, need for control, and subsequent acceptance of advanced
assistance systems. Through simulator experiments, in-depth qualitative
interviews, as well as situation awareness tests, we have sought to understand
how in situations of varying criticality, drivers respond in terms of their
acceptance of risk, as well as assistance. The uniqueness of this study lies in
the depth of understanding it provides on driver motivations by adopting a
qualitative methodology and the analytical tool of the activity framework.
Through this lens we have seen how shifting perceptions on risk and control
determine the efficacy and acceptance of ADAS systems. We have examined
tangents in the activity system that result in outcomes not always matching the
desired object, as in this case we saw with risk-compensation. In terms of future
directions for this research, we aim at continuing our analysis efforts both in
terms of driver diversity in risk-taking, as well as in terms of user acceptance of
ADAS. Parallel studies that were conducted using video tools and focussing on
sensation-seekers and risk takers, will be integrated with the findings of this
project at a wider level.
6 References
[1]
[2]
[3]
[4]
[5]
[6]
[7]
[8]
[9]
Deery, H.A., 1999, Hazard and risk perception among young novice
drivers. Journal of Safety Research .30(4), pp 225–236
Van der Laan, Heino, and De Waard’s (1997), A simple procedure for the
acceptance of assistance of advanced transport telematics. Transportation
Research, C 5 (1), 1-1
Heijden, R.E.C.M. van der, Wiethoff, M. (eds, 2000; Eds) Automation of
Car driving. Exploring societal impacts and conditions. Delft: DUP. ISBN
90-407-2028-2
Vygotsky, L.S. (1978). Mind in Society. Cambridge, MA: Harvard University
Press
Engeström, Y. (1987). Learning by expanding, Helsinki: Orienta-konsultit
Burger, J. M. (1986). Desire for control and the illusion of control: The
effects of familiarity and sequence of outcomes. Journal of Research in
Personality, 20, pp 66–76
Rosengren, L.G. (1995). Driver assistance and co-operative driving. In:
ERTICO (Ed.), Towards an intelligent transport system. Proceedings of the
First World Congress on Advanced Transport Telematics and Intelligent
Vehicle Highway Systems. London: Artech House, pp 1613-1622
Zwahlen, H.T., Adams, C.C. & DeBald, D.P. (1988). Safety aspects of CRT
touch panel controls in automobiles. In: A.G. Gale, M.H. Freeman, C.M.
Haslegrave, P. Smith & S.H. Taylor (Eds.). Vision in Vehicles - II.
Amsterdam: Elsevier, North-Holland, pp 335-344
Verwey, W.B., Brookhuis, K.A., Janssen, W.H. (1996). Safety effects of invehicle information systems. Report TM-96-C002. Soesterberg, the
Netherlands: TNO Human Factors Research Institute
344
Drivers’ acceptance of assistance functions
[10] Janssen, W.H., The Effect of Seat Belt Legislation on Fatality Rates in a
Number of West European Countries, Report Number 1ZF, 1989-13, TNO
Institute for Perception, Soesterberg, Netherlands
[11] Bekiaris, E., Petica, S., Brookhuis, K.A. (1997). Driver needs and public
acceptance regarding telematic in-vehicle emergency control aids. In:
Proceedings of the 4th Word Congress on Intelligent Transport Systems.
Brussel: Ertico (on CD)
[12] Nilsson, L. & Alm, H. (1991). Collision Avoidance Systems -Effects of
different levels of task allocation on driver behaviour Report DRIVE
V1041/GIDS/MAN3. Haren, The Netherlands: University of Groningen,
Traffic Research Centre
[13] Hoedemaeker, M. (1996). Needs, driving styles and opinions towards
Automated Driving. Report. Delft, The Netherlands, Technical University
Delft
[14] Comte, S., Wardman, M., Whelan, G., (2000), Driver’ acceptance of
automatic speed limiters: implications for policy and implementation.
Transport Policy 7, pp 259 - 267
345
Drivers’ acceptance of assistance functions
THE EFFECT OF ISA IN RELATION TO SPEED
LIMIT CREDIBILITY
Nicole van Nes, Ingrid van Schagen (SWOV, The Netherlands)
Nicole.van.Nes@SWOV.nl
ABSTRACT: One of the most challenging areas for ITS to contribute to road
safety is controlling speed behaviour. Speeding is still common practise on
many roads and contributes to a significant number of crashes. An Intelligent
Speed Assistance (ISA) aims to reduce speeding in a more or less intrusive
manner. This is promising, however the acceptance of ISA systems is still very
low. Another promising approach to reduce speeding is by improving the
credibility of speed limits. A driving simulator study is conducted to investigate
the effect of credibility, with or without using an open informative and warning
ISA. The results show that both ISA and credibility of speed limits are effective
measures to reduce speeding. Non-ISA users appear to be more sensitive for
the credibility of speed limits than ISA users. With regard to the acceptance of
the ISA system used in this study, the expectations were fairly positive than the
experiences.
1 Introduction
1.1
Background
The faster one drives, the more likely a crash, and, in case of a crash, the more
severe the injury consequences [1,2]. Excessive speed is a very common
phenomenon in Europe and many non-European countries. Typically, at any
time 50% of drivers are exceeding the speed limit [3] and a large safety benefit
could be obtained if driving speeds were reduced. There are a number
traditional and well proven measures to reduce driving speeds and to make
drivers comply better with the speed limit, in particular infrastructure measures
and speed enforcement. An effect is also assumed from credible speed limits,
i.e. speed limits that meet the expectations of the road users, given the road
features and the road surroundings. Credibility will not only have an effect in
itself, but may also enhance the effects and the acceptance of other types of
measures [3, 4]
A newer, and very promising measure to reduce speeds and the number of
speed-related crashes is Intelligent Speed Assistance (ISA). It has been
calculated, based on the known relationship between speed and crash
involvement/severity, that ISA will have a substantial effect on road safety, in
particular the more intervening systems that make speeding physically
impossible and that are based on dynamic speed limits [5].
While such advanced systems may be technically possible, their widespread
use will need more time. Less advanced systems, based on fixed speed limits,
have been tested in various countries, both in small and large field trials,
Human Centred Design for Intelligent Transport Systems
generally with positive effects on average speed [e.g. 6, 7, 8, 9]. Less
intervening systems, those only informing the driver about the (fixed) speed limit
in force are already on the market, e.g. Speedalert.
One of the important issues that hinder the widespread implementation of ISA is
related to the public acceptance of systems. While more than half of the
European drivers state that they are in favour of a system that prevents them
from exceeding the speed limit [10], drivers participating in field trials had less
positive ideas about ISA. However, after they had experienced the system, their
opinion turned out to be more positive [6, 9].
1.2
Aim and research questions
This study aimed to investigate whether credibility affects speed choice when
driving with an ISA system that informs the driver about the speed limit in force
and warns if the speed limit is exceeded. As public acceptance is of major
importance, the acceptance of the used ISA system was also investigated. This
lead into the following research questions:
1. What is the effect of speed limit credibility on speed choice for ISA users in
comparison to non-ISA users?
2. What is the difference in acceptance of the ISA system before and after the
(test drive) experience?
2 Method
2.1
Design of the study
The research questions were investigated in a driving simulator study. A total of
41 subjects drove along a simulated network of mainly rural roads with speed
limits of 60, 80 or 100 km/h. The study focused on speed choice at 7 road
sections with a speed limit of 80 km/h, the other road sections were added to
create a realistic environment.
2.2
Independent variables
In the experiment two variables were manipulated, the use of ISA and the
credibility of the speed limit. The use of ISA was varied between subjects, half
of the subjects were supported by an ISA system, and the other half were not.
The ISA system used in this experiment was integrated in a navigation device.
The ISA system provided continuous visual information about the speed limit in
force and when the driver exceeded the speed limit the ISA system warned the
driver visually (the speed limit indicator on the navigation device increased in
size and flickered) and verbally (a female voice said: "You're exceeding the
speed limit. The speed limit is .. km/h." This message was repeated every ten
seconds until the speed was reduced to below the speed limit). Exceeding the
speed limit was defined as 1 km/h or more over the posted limit.
The credibility of the speed limit was manipulated by varying a number of road
characteristics that were identified in literature as being relevant for speed
choice [e.g. 11, 12, 13], and that were possible to simulate in a driving
348
Drivers’ acceptance of assistance functions
simulator. Road bendiness for example is considered as relevant for speed
choice, but hard to simulate in a driving simulator because it increases the
chance of simulator sickness. The road characteristics that were used to
manipulate the credibility were road width, presence of vegetation near the
road, and the type of separation between the driving directions. Each road
characteristic was expected to have an effect on the intuitive speed, like a small
road was assumed to reduce speed while a wide road was assumed to increase
speed. Each road section had a different combination of characteristics. At
some road sections all of the manipulated road characteristics were assumed to
increase the intuitive speed, on the other road sections a mixture of
characteristics that were assumed to increase speed and those that were
assumed to decrease speed. Based on this the road sections were categorised
as having a more or less credible speed limit. As literature on the subject was
quite limited, the credibility of the speed limit was validated within the
experiment. All subjects first drove the test route without showing any speed
limit. The results showed that at the 80 km/h roads the subjects intuitively drove
about 80 km/h at the road sections with credible speed limits, and that they
exceeded the speed limit by more than 5% at the road sections with less
credible speed limits.
2.3
Dependent variables
2.3.1 Driving speed
The main dependent variable was driving speed. To measure driving speed two
variables were measured: average driving speed and speeding time. The
average driving speed was computed for each road section. To avoid an effect
of acceleration and deceleration near intersections, the computed averages are
based on the part of the road section from 100 m after an intersection until 150
m before the next intersection. Speeding time was defined as the percentage of
time on a road section that the limit was exceeded by 10% or more, thus 88
km/h or more at 80 km/h roads.
2.3.2 Acceptance of the ISA system
The ISA users were informed about the ISA system before the test drive. They
received a paper with information on the system and the type of information it
would provide. Based on this information the subjects were asked about their
expectations of the system. After the test drive, they were asked about their
satisfaction with the ISA system based on their actual experience. The
expectations and experiences were measured by presenting statements and a
five-point answering scale (completely disagree, disagree, neutral, agree,
completely agree). The same set of statements was used for the expectations
and for the experiences, the only difference was the tense, the first set was in
the present tense while the second set was in the past tense. (E.g. "It seems
pleasant to me that…." versus "I found it pleasant that ….").
2.4
Analysis
The effects of ISA and the credibility of speed limits were determined by means
of an analysis of variance with repeated measurements. Within these analyses
the use of ISA was considered as a 'between subject factor'. The repeated
Human Centred Design for Intelligent Transport Systems
measurement concerned the speed behaviour on the different road sections
with different speed limit credibility. For all analyses, a critical significance level
of 5% was applied.
3 Results
The results for average driving speed and speeding time are shown in
respectively Figure 1 and Figure 2. In the figures the values are shown for ISA
users and non-ISA users at road sections with low and high credibility of the
limits.
30,00
non-ISA user
ISA user
25,00
82,50
Speeding time (%)
Driving speed (km/h)
non--ISA user
ISA user
80,00
77,50
20,00
15,00
10,00
5,00
0,00
low
low
high
high
Credibility
Credibility
Fig.1. Average driving speedFig.2. Speeding time
3.1
The effect of ISA
The main effect of ISA on driving speed was significant (F1,39=11.33; p<0.01).
ISA users drove significantly slower than non ISA users. Moreover, ISA users
on average had a driving speed below the speed limit (77.5 km/h), whereas
non-ISA users had an average driving speed above the speed limit (82.3 km/h).
The main effect of ISA on speeding time was also significant (F1,39=10.76;
p<0.01). ISA users exceeded the speed limit significantly less than non ISA
users. ISA users, on average, did not exceed the speed limit at road sections
with credible speed limits and hardly exceeded the speed limit at road sections
with less credible speed limits.
3.2
The effect of credibility
The main effect of speed limit credibility on average driving speed was
significant (F1,39=19.61; p<0.001). The subjects drove significantly slower at
road sections with credible speed limits than at road sections with less credible
speed limits. This is the effect for the complete sample, half of the subjects with
and half without ISA. As the current practise is that hardly anyone uses an ISA
system, it is also interesting to look at the effect for non-ISA users only. A
simple main effects analysis showed that for non-ISA users the effect of
350
Drivers’ acceptance of assistance functions
credibility was significant (F1,39=18.38; p<0.001). Average driving speed was
lower at sections with credible limits than at sections with less credible sections.
For the ISA users the effect of credibility was not significant, but there is a trend
(F1,39=3.79; p<0.1) that also ISA users drive slower at sections with credible
limits.
The main effect of the credibility of speed limits on speeding time was also
significant (F1,39=12.41; p<0.005). The subjects exceeded the speed limit
significantly less at road sections with credible speed limits than at road
sections with less credible speed limits. Again, when looking separately at the
ISA and the non-ISA users, the effect of credibility was significant for the nonuser group (F1,39=17.08; p<0.001), but there was only a trend for the ISA users
(F1,39=0.66; p<0.5).
3.3
The interaction between ISA and credibility
The interaction effect between ISA and credibility on average driving speed was
not significant, but there is a trend (F1,39=2.92; p<0.1) that non-ISA users were
more sensitive for the credibility of speed limits than ISA users. This trend is
supported by the simple main effects reported in the previous section that
indicate that the effect of credibility is significant for non-ISA users and not for
ISA users.
The interaction effect between ISA and credibility on speeding time was
significant (F1,39=5.72; p<0.05). Non-ISA users were more sensitive for the
credibility of speed limits than ISA users.
3.4
Acceptance of the ISA system
The responses to the statements on the expectations of and experiences with
the ISA system are presented in Figure 3. The results show that expectations of
the subjects were somewhat positive and their satisfaction slightly decreased to
neutral after the test. In particular the voice that warned when exceeding the
speed was not appreciated. The experience did not change much of their
opinion on the visual information and the visual warning, they were well
accepted.
Human Centred Design for Intelligent Transport Systems
agree2
completely
expectations
experiences
agree1
neutral0
disagree-1
ISA is com fortable visual inform ation is
(overall perception)
com fortable
visual w arning is
com fortable
voice w arning is
com fortable
Fig.3. Expectations of and experience with the ISA system
4 Conclusions and discussion
4.1
Effect of ISA
The study found a significant effect of ISA on the average driving speed and on
speeding time. The ISA system in the study that both informed and warned
when the speed limit was exceeded resulted in lower speeds and less time
driving above the speed limit. These results confirm the findings of the large
majority of studies that investigated the effect of ISA [ e.g.6, 9, 14, 15,16].
4.2
The effect of credibility
The results indicated that, overall, credible speed limits result in lower speeds
and less time exceeding the speed limit than speed limits that are less credible,
because they are considered to be too low. Systematic research on the effects
of credible speed limits is still lacking, but these results confirm the assumptions
made by many experts [e.g. 3, 4, 17, 18].
4.3
The effect of ISA related to speed limit credibility
The first main research question was whether the effect of credible speed limits
was similar for ISA and non-ISA users. When looking at the time drivers
speeded, this appeared not to be the case. The effect of speed limit credibility
was stronger for non-ISA users than for ISA users. Looking at average speed,
however, there was no significant difference between ISA users and non-ISA
users, although the trend pointed in the same direction. Looking separately at
the effects of credibility for ISA and non-ISA users, it appeared that this had an
effect without ISA, but not with ISA drivers, both for average speed and for
speeding. So, the results of this study suggest that, when driving with an
informing and warning type of ISA, the credibility of speed limits does not affect
352
Drivers’ acceptance of assistance functions
the amount of time people drive in excess of the speed limit. Regarding average
speed, the results are less clear, but point in the same direction.
When interpreting this outcome, it must be born in mind that the ISA used in this
study was an open one. Drivers could make their own choice about their driving
speed. However, within the open category, it was a fairly intrusive system.
Warnings started when the speed limit was exceeded by just 1 km/h and were
repeated every 10 seconds, not only visual but also verbally. It would be well
possible that speed limit credibility would affect driving speed of ISA drivers
more, when the system was only informing or when the warnings were less
frequent and less intrusive.
4.4
Acceptance of the ISA system
The second main research question was related to the acceptance of the ISAsystem by the drivers. The results showed that the subjects had fairly positive
expectations. However, they were somewhat disappointed after they had
actually experienced it. In particular, the overall convenience of this type of ISA
and of the verbal warning was judged less positive. The verbal warning was
even judge negatively.
This finding is in line with another study with an informative ISA system that also
concludes that there is no difference in acceptance before and after a test ride.
However, most studies report that drivers become more enthusiastic about ISA,
once they have experienced it in practice [6, 9, 19, 20].
One explanation of our finding may be that this particular ISA system was quite
strict, with visual and verbal warnings at 1 km/h or more over the limit and
repeated every 10 seconds. This could particularly explain why the drivers
disliked the verbal warning system. Acceptance of the system may increase if
the threshold for warnings would be higher and the warning frequency lower.
Other studies also found that the less intervening systems are accepted better
[5]. At the same time it is expected that, in that case, the effect of the ISA
system would be smaller. Clearly, it is a matter of finding a balance between
acceptance and effectiveness of the system.
Another explanation of the contradictory findings in our study may lie in the fact
that our study focused on rural roads and more particularly on 80 km/h roads. It
is well possible that the acceptance of ISA is different for urban roads (with 30
and 50 km/h speed limits) than for rural roads.
Our study did not make a distinction between the acceptance of ISA at road
sections with a credible speed limit and sections where the speed limit was
considered to be too low. Hence, based on the current study, we cannot say
whether credible speed limits improve acceptance.
4.5
Overall conclusion
Based on the results of this study we can conclude that ISA is a promising
instrument to reduce speeding and increase road safety. In line with previous
research, the results of this study indicate that an (open) ISA system on its own
has the potential to reduce average driving speed to below the speed limit and
to decrease speeding.
Human Centred Design for Intelligent Transport Systems
Credibility of speed limits is also an important factor in speed choice and can
substantially contribute to achieving lower speeds and less speeding. Still,
credible speed limits cannot achieve the same result as the informative and
warning ISA system of the current study.
Speed limit credibility seems to have hardly an effect if speed choice is
supported by a rather strict informative and warning ISA system. It is expected,
however, that speed limit credibility will be more important in case of less
intruding systems that can be more easily ignored. In addition, it is likely that
credible speed limits will increase acceptance of ISA systems. Future research
would need to find out.
It is important to be aware of the fact that speeding behaviour is related to many
different factors that are hard to realize in a driving simulator study, like time
pressure, whether the driver expects to be punished for speeding, risk
estimation and so on. In this study the some influencing factors, ISA and
credibility, were studied. For future research it would be interesting to do a
similar study on the real road in order to validate if the results are the same and
it would be interesting to look and the interaction of the effects investigated in
this study and other factors that effect speeding behaviour.
5 References
[1]
[2]
[3]
[4]
[5]
[6]
[7]
[8]
Aarts, L. & Van Schagen, I.N.L.G. Driving speed and the risk of road
crashes; A review. Accident Analysis and Prevention, 2006, (2), pp 215224.
Elvik, R., Christensen, P. & Amundsen, A. Speed and road accidents; An
evaluation of the Power Model. Institute of Transport Economics TØI, Oslo,
2004.
OECD/ECMT. Speed management. Organisation for Economic Cooperation and Development OECD/European Conference of Ministers of
Transport ECMT, Paris, 2006.
Wegman, F & Aarts, L. (Eds.) Advancing sustainable safety. SWOV
Institute for Road Safety Research, Leidschendam, 2006.
Carsten, O.M.J. & Tate F.N . Intelligent speed adaptation: accident savings
and cost–benefit analysis. Accident Analysis & Prevention, 2005, 37(3),
pp 407-416.
Biding, T. & Lind, G. Intelligent Speed Adaptation (ISA); Results of largescale trials in Borlänge, Lidköping, Lund and Umeå during 1999-2002.
Publication 2002:89 E. Swedish National Road Administration SNRA,
Borlänge, 2002.
Vlassenroot, S. Broekx, S., De Mol, J., Int Panis, L., Brijs, T., Wets, G.
Driving with intelligent speed adaptation: Final results of the Belgian ISAtrial. Transportation Research A, 2007, 41(3), pp 267-279.
Regan, M.A. Thomas J. Triggs, Kristie L. Young, Nebojsa Tomasevic, Eve
Mitsopoulos, Karen Stephan, & Claes Tingvall. On-road evaluation of
Intelligent Speed Adaptation, Following Distance Warning and Seatbelt
Reminder Systems: Final results of the TAC SafeCar project. Monash
University Accident Research Centre, Melborn, 2006.
354
Drivers’ acceptance of assistance functions
[9]
[10]
[11]
[12]
[13]
[14]
[15]
[16]
[17]
[18]
[19]
[20]
Besseling, H. & Boxtel, A. van Intelligent Speed Adaptation: results of the
Dutch ISA Tilburg trial: summary. Ministry of Transport, Public Works and
Water Management, Transport Research Centre AVV, Rotterdam, 2001.
Cauzard, J.-P. (ed.) European drivers and road risk: Project on Social
Attitudes to Road Traffic Risk in Europe SARTRE 3. Institut National de
Recherche sur les Transports et leur Sécurité INRETS, Arcueil, 2004.
Goldenbeld, C. and I. van Schagen, The credibility of speed limits on 80
km/h rural roads: The effects of road and person(ality) characteristics.
Accident Analysis & Prevention, 2007, 39(6): pp. 1121-1130.
Elliot, M.A., McColl, V.A. & Kennedy, J.V. Road design measures to
reduce drivers' speed via 'psychological' processes: a literature review.
TRL report TRL564. Transport Research Laboratory, Crowthorne, 2003.
Martens, M., Comte, S. & Kaptein, N. The effects of road design on speed
behaviour; A literature review. Deliverable 1 of the MASTER project. TNO
report TM-97-B021. TNO Human Factors, Soesterberg, 1997.
Madsen, J.R. Intelligent Speed Adaptation. An analysis of behavioural
changes. In: E-safety: IT Solutions for Safety and Security in Intelligent
Transport. International Congress & Exhibition, 16-18 September 2002,
Lyon, France. ERTICO, European Commission, ITS France, 2002.
Persson, H., Towliat, M., Almqvist, S., Risser, R. & Magdebuirg, M. Speed
limiters for cars. A field study of driving speeds, driver behaviour, traffic
conflicts and comments by drivers in town and city traffic. Department of
Traffic Planning and Engineering, University of Lund, Lund, Sweden, 1993.
Várhelyi, A. & Mäkinen, T. The effects of in-car speed limiters: Field study.
Transportation Research, Part C., 2001, pp. 191-211.
Risser R. & Lehner, U. Acceptability of speeds and speed limits to drivers
and pedestrians/cyclists. Master Deliverable D6. CEC DG-TREN, Brussel,
1998.
Fildes, B.N. & Lee, S.J. The speed review: road environment, behaviour,
speed limits, enforcement and crashes. Report CR 127. Federal Office of
Road Safety; Department of Transport and Communication, Canberra,
Australië, 1993.
Marchau, V.A.W.J., Heijden, R.E.C.M. van der & Molin, E.J.E. Societal
needs regarding ADAS implementation: the case of ISA. In: E-safety: IT
Solutions for Safety and Security in Intelligent Transport. International
Congress & Exhibition, 16-18 September 2002, Lyon, France. ERTICO,
European Commission, ITS France.
Wiethoff, M., Oei, H.L., Penttinen, M., Anttila, V. & Marchau, V.A.W.J.
Advanced Driver Assistance Systems: an overview and actor position. In:
Proceedings of the 15th Triennial IFAC World Conference, July 2002,
Barcelona.
INDEX OF AUTHORS
Aillerie ..........................................89
Häger ........................................ 175
Akamatsu.................................. 121
Hamada....................................... 13
Amditis ...................................... 335
Hamama.................................... 239
Auriault.........................................89
Hannibal .................................... 175
Banet ........................................ 287
Henning............................. 113, 277
Barbé ........................................ 183
Hoc............................................ 311
Baumann .......................................3
Hogema......................................... 3
Bekiaris ..................................... 217
Joshi.......................................... 335
Bellet......................... 277, 287, 335
Karlaftis ..................................... 163
Bodard ...................................... 335
Kassner ............................. 193, 327
Boets......................................... 217
Kaufmann............................ 33, 101
Bonnard .......................................43
Krems.................... 3, 113, 277, 301
Boy............................................ 183
Krüger ....................................... 153
Brookhuis.....................................23
Kullack....................................... 321
Brusque ................................ 43, 89
Leden ........................................ 253
Bruyas................................. 89, 239
Mars .......................................... 311
Cotter ...........................................55
Mayora ...................................... 205
Duraz ...........................................89
Navarro ..................................... 311
Eggert ....................................... 321
Nilsson ...................................... 301
Ehrenpfordt ............................... 321
Pauzié ......................................... 67
El Jaafari................................... 311
Per Henriksson ......................... 217
Falkenstein ..................................23
Pereira....................................... 239
Forzy......................................... 311
Petzoldt ......................................... 3
Gačnik....................................... 175
Popken ................................ 55, 301
Gelau ...........................................55
Portouli ...................................... 217
Georgeon.......................... 113, 277
Rauch........................................ 153
Geven ..........................................33
Richardson .......................... 79, 277
Gradenegger............................. 153
Rinkenauer.................................. 23
Groenewoud ..................................3
Risser .......................................... 33
Human Centred Design for Intelligent Transport Systems
Sato........................................... 121
Turetschek ........................ 101, 265
Schaap...................................... 131
van Arem................................... 131
Schießl ...................................... 141
van der Horst............................. 131
Schmidt ..................................... 229
van Nes ..................................... 347
Sefelin ......................................... 33
van Schagen ............................. 347
Simões ...................................... 239
Vollrath ...................................... 193
Spyropoulou.............................. 163
Werneke.................................... 193
Stevens ....................................... 55
Wilschut....................................... 23
Tattegrain.................................... 89
Wynn ........................... 79, 113, 277
358
Olivier Georgeon