Transportation Research Part F 10 (2007) 139–152
www.elsevier.com/locate/trf
Impacts of a Congestion Assistant on driving behaviour
and acceptance using a driving simulator
C.J.G. van Driel
a
a,*
, M. Hoedemaeker b, B. van Arem
a
University of Twente, Centre for Transport Studies, Applications of Integrated Driver Assistance, P.O. Box 217,
7500 AE Enschede, The Netherlands
b
TNO Human Factors, P.O. Box 23, 3769 ZG Soesterberg, The Netherlands
Received 24 May 2006; received in revised form 14 August 2006; accepted 15 August 2006
Abstract
In-vehicle systems that assist the driver with his driving task are developed and introduced to the market at increasing
rate. Drivers may be supported during congested traffic conditions by a so-called Congestion Assistant consisting of a mix
of informing, assisting and controlling functions. This paper describes the impacts of the Congestion Assistant on the driver in terms of driving behaviour and acceptance. Thirty-seven participants took part in a driving simulator study. The
observed driving behaviour showed promising improvements in traffic safety when approaching the traffic jam. Moreover,
positive effects of the system on traffic efficiency can be expected in the jam. The participants stated to appreciate the Congestion Assistant, although not all functions were equally rated. To increase the performance and acceptance of the total
system, some refinements were suggested.
2006 Elsevier Ltd. All rights reserved.
Keywords: Congestion Assistant; Driving simulator; Integrated system; Driving behaviour; Acceptance
1. Introduction
Over the coming years, drivers will have an increasing variety of so-called Advanced Driver Assistance
(ADA) systems at their disposal. These are in-vehicle systems that support the driver with the driving task.
ADA systems are expected to lead to a safer, cleaner and more efficient and comfortable transport system
(European Commission, 2002; Ministry of Transport, 2004). However, the extent to which ADA systems will
meet these high expectations is greatly dependent on the willingness of car drivers to use these systems. To gain
more insight into the perceived needs for driver assistance, a survey among 1049 Dutch motorists was conducted (Van Driel & Van Arem, 2005). It appeared that there was a significant need for warnings of downstream traffic conditions (e.g., congestion, road works). Moreover, the ideal driver support system should
support the driver in critical situations, such as an imminent crash or reduced visibility, and with congestion
*
Corresponding author. Tel.: +31 53 4894503; fax: +31 53 4894040.
E-mail address: [email protected] (C.J.G. van Driel).
1369-8478/$ - see front matter 2006 Elsevier Ltd. All rights reserved.
doi:10.1016/j.trf.2006.08.003
140
C.J.G. van Driel et al. / Transportation Research Part F 10 (2007) 139–152
driving on motorways. Apparently, drivers appreciated several forms of congestion assistance. The survey
results served as a basis for designing the so-called Congestion Assistant. This integrated in-vehicle system
consisted of a mix of informing, assisting and controlling functions to support the driver during congested
traffic situations on motorways.
1.1. Previous research into congestion assistance
Congestion can be considered a major problem of modern societies. On a European scale, some 7500 km or
10% of the road network is affected daily by traffic jams (European Commission, 2001). Therefore, research is
focused on mitigating the effects of congestion. Among others, ADA systems can provide congestion assistance. Consider, for example, the extension of Adaptive Cruise Control (ACC) with Stop & Go to reduce
the driver’s burden of congestion driving (Marsden, Brackstone, & McDonald, 2001; Venhovens, Naab, &
Adiprasito, 2000). This ADA system is capable of both regulating speed and following distance, even in congested traffic conditions (Delphi, 2005). Furthermore, the Traffic Congestion Assistant of INVENT aims to
support the driver in the low speed segment in non-urban congested areas by giving longitudinal and lateral
control (Hummel et al., 2003). In a traffic simulation study, this system showed an increased traffic flow and a
decreased fuel consumption when operating at a time headway of 1.0 s or 1.8 s (Benz, Christen, Lerner, &
Vollmer, 2003). Besides autonomous systems, research is also focused on developing co-operative systems that
are based on inter-vehicle or vehicle–roadside communication. For example, the Traffic Performance Assistance of INVENT is designed to make the traffic flow more smoothly and relieve traffic jams (Krautter,
Mackamul, Manga, & Manstetten, 2004). Results of a driving simulator experiment showed that with an
informative version of this system the participants reduced their speed earlier when approaching a traffic
jam and drove more smoothly in the jam. Applications in the CarTALK 2000 project used inter-vehicle communication to broadcast a message to upstream vehicles when accelerating at 2.0 m/s2 or less (Malone &
Van Arem, 2004). The results of a traffic simulation study indicated an improvement of traffic stability in
terms of a reduction in the number of shockwaves for all penetration rates and headways tested.
1.2. Assessment of ADA systems
Before introducing ADA systems on the market, a clear understanding of the impacts of these systems on
the driver and the traffic system should be available. Impact assessment involves the measurement or estimation of the impacts of an application, for example on user behaviour, traffic safety or traffic efficiency. User
acceptance assessment aims to estimate users’ attitudes to and perception of an application, usually based
on questionnaire surveys or interviews.
Measures of driving behaviour are used for a wide range of applications. In general, the assessment is concerned with determining the potential driving performance enhancements resulting from the support given by
the system and decrements resulting from the dual task of using the system while driving. Many measures of
driving behaviour exist, which particularly score the performance of the tactical and operational levels of the
driving task (Michon, 1985). The goal of an assessment is often to identify any effects of the ADA system.
Therefore, generally a large set of measures is used. The HASTE project developed a test regime for the assessment of in-vehicle information systems (e.g., navigation system) (Johansson et al., 2005). The test regime consisted of safety-related measures that can indicate reduced driving performance due to visual and/or cognitive
loads imposed by in-vehicle information systems. It was concluded that the following measures should be
taken into account for the evaluation of these systems: mean speed, proportion of high frequency steering
activity, minimum time headway and subjective rating of the quality of one’s own driving performance.
A prerequisite for the introduction of new in-vehicle technology is acceptance by the public (Van der Laan,
Heino, & De Waard, 1997). It would be unproductive to invest effort in designing and building ADA systems
which will never be purchased and used. Regan, Mitsopoulos, Haworth, and Young (2002) stated that usefulness, ease of use, effectiveness, affordability and social acceptability are the key constructs that are assumed to
underlie the concept of acceptability. Van der Laan et al. (1997) developed a standardised checklist for the
assessment of acceptance of new in-vehicle technology. It consists of nine 5-point rating-scale items that indicate a usefulness scale and a satisfaction scale. The checklist appeared to be sensitive to differences of opinion
C.J.G. van Driel et al. / Transportation Research Part F 10 (2007) 139–152
141
with respect to specific aspects of in-vehicle systems, as well as to differences of opinion among different driver
groups.
1.3. The present study
The main objective of the present study was to assess the impacts of the Congestion Assistant on the driver
in terms of driving behaviour and acceptance. For this reason, a driving simulator experiment was carried out.
Knowledge of the system’s impacts on the driver will be used for a traffic simulation study to gain more insight
into the system’s impacts on a whole traffic flow in terms of traffic safety and efficiency. For the present study,
it was hypothesized that drivers with the Congestion Assistant would display safer and more efficient driving
behaviour than drivers without the system. Anticipated effects of the informing and assisting functions of the
system include earlier and smoother decelerations by the drivers when approaching the traffic jam, while the
controlling function of the system could perform the car-following task in the traffic jam more efficiently (i.e.,
follow at closer headways with less variation). Respondents to the user needs survey indicated strong needs for
support in reduced visibility conditions (Van Driel & Van Arem, 2005). Therefore, it was expected that the
Congestion Assistant would be appreciated more during such situations (e.g., in fog) than during clear sight.
To validate the survey results, respondents were asked to participate in this driving simulator experiment. In
this way, the attitude towards congestion assistance – based on the survey results – could be related to the
acceptance of the Congestion Assistant. It was hypothesized that participants with a positive attitude towards
congestion assistance in general would also be more positive about the Congestion Assistant than participants
with a negative attitude.
2. Method
2.1. Congestion Assistant
The Congestion Assistant supported the driver during congested traffic situations on motorways. It consisted of the following functions: (1) congestion warning and information: while driving towards the traffic
jam, the driver received a warning of the traffic jam ahead. While driving in the traffic jam, information about
the length of the traffic jam was displayed, (2) active gas pedal: while approaching the traffic jam, the driver
could feel a counterforce on the gas pedal when the speed was too high according to the system and (3) Stop &
Go: while driving in the traffic jam, the longitudinal driving task was taken over by the system.
The Congestion Assistant gave the driver a warning when he was approaching a traffic jam. It was expected
that the driver would be better prepared for the traffic conditions ahead, which could be expressed by earlier
and smoother decelerations. The congestion warning was presented on a display, which was mounted on the
centre console (Fig. 1). The first congestion warning was introduced by a sound signal and a corresponding
icon lighting up. The warning consisted of a text message informing the driver about the distance and time
he was apart from the traffic jam. This information was updated every half kilometre. The congestion warning
was active from 5 km before the traffic jam until the tail of the traffic jam. Furthermore, the Congestion Assistant provided the driver with information when he was driving in the traffic jam. The congestion information
was presented on the display, while the corresponding icon was still lightened up. The information consisted of
a text message informing the driver about the remaining distance of the traffic jam. This information was
updated every half kilometre.
When the driver approached the traffic jam at too high speed, the active gas pedal of the Congestion Assistant gave him a warning by means of a counterforce on the gas pedal (maximum 50 N). It was expected that
the driver would better anticipate the traffic jam by earlier and smoother decelerations and by safer car-following behaviour. The active gas pedal was working from 1.5 km before the traffic jam until the tail of the
traffic jam. The activation was introduced by a sound signal and a corresponding icon lighting up on the display. The principle of the active gas pedal was similar to that of the Intelligent Speed Adaptation system
described by Hogema and Rook (2004). The Congestion Assistant worked with a reference acceleration that
represented the necessary acceleration (deceleration) for safely approaching the traffic jam. The reference
acceleration was calculated based on the distance from the tail of the jam, the actual speed and the speed
142
C.J.G. van Driel et al. / Transportation Research Part F 10 (2007) 139–152
Fig. 1. Display of the Congestion Assistant on the centre console.
of vehicles in the tail of the jam which was set at 50 km/h. The active gas pedal gave a counterforce when the
reference acceleration was below the threshold of .5 m/s2. This threshold implied that people driving at
120 km/h would feel a counterforce on the gas pedal around 900 m from the tail of the traffic jam. When
the tail was reached, the active gas pedal became inactive and the corresponding icon turned off.
In the traffic jam, the Stop & Go function of the Congestion Assistant took over the longitudinal driving
task (i.e., regulating speed, car-following, accelerating after standstill). For safety reasons, the driver had to
perform the lateral driving task himself to keep him involved in car driving (i.e., ‘driver in the loop’) (Stanton
& Young, 1998). The activation of the Stop & Go was introduced by a spoken message ‘‘The Stop & Go will
turn on’’, a sound signal and a corresponding icon lighting up on the display. At the end of the traffic jam, the
Stop & Go and congestion information were deactivated. This was again introduced by a spoken message
‘‘The Stop & Go will turn off’’. Next, a sound signal was presented and the corresponding icons were turned
off. The driver had to take over from the Stop & Go and perform the longitudinal task himself again. About
2 s after the spoken messages, the Stop & Go actually became (in)active. Preparing the driver for each transition between functions of the Congestion Assistant by sound signals, spoken messages and icons was
expected to reduce the negative effects of so-called ‘task switching’ (e.g., increase of reaction time) (Monsell,
2003). The features of the Stop & Go are given in Table 1. It was expected that the Stop & Go would perform
‘better’ than the driver when driving in stop-and-go traffic. For example, the Stop & Go might better anticipate leading vehicles and thus accelerate and decelerate in a smoother way. Also, the Stop & Go could lead to
car-following at closer headways with less variation.
It was assumed that the Congestion Assistant received information about the state of the traffic flow (e.g.,
by means of inter-vehicle and vehicle–roadside communication), so that it could provide the driver with the
appropriate support function. However, to simplify things in this driving simulator experiment, a static traffic
jam with a fixed start and end was created. The functions of the Congestion Assistant were programmed to
Table 1
Main features of the Stop & Go
Feature
Stop & Go
Speed range
Reference speed
Reference time headway
Minimum distance headway (at standstill)
Maximum acceleration
Maximum deceleration
Automatic ‘‘go’’
Overrulable
Sensor range
0–70 km/h
70 km/h
1.0 s
3m
2 m/s2
9 m/s2
Yes
No
200 m
C.J.G. van Driel et al. / Transportation Research Part F 10 (2007) 139–152
143
Fig. 2. Schedule of visual and acoustic feedback by the Congestion Assistant.
switch on or off at established points during the experimental runs (see ‘sound signals’ in Fig. 2). The driver
was not able to (de)activate (functions of) the Congestion Assistant himself.
2.2. Research design
A repeated measures design with System (without versus with Congestion Assistant) and Visibility (normal
versus fog) as within-subject factors was used. To avoid order and learning effects, the conditions were counterbalanced by a Latin square design.
2.3. Scenarios
The test road environment consisted of a motorway with a speed limit of 120 km/h. The motorway was
based on 2 · 2 traffic lanes and shoulders according to the Dutch design guidelines. The road was mainly
straight with a few gentle curves. Each participant encountered a traffic jam during an experimental run.
All surrounding vehicles were ‘told’ at what speed they had to drive (e.g., 50 km/h in the traffic jam,
120 km/h outside the traffic jam).
For the fog condition, a constant visibility setting of 80 m was used, which can be regarded as dense fog.
When driving in fog, drivers behave differently than during driving in clear sight. To come close to reality, the
behaviour of the other simulated traffic was altered in line with the findings of Hogema and Van der Horst
(1994). For example, it was assumed that drivers of a passenger car would drive at a speed of approximately
100 km/h in fog (in contrast to 120 km/h during normal visibility).
Each run had a length of 15 km and was divided into the following five traffic scenarios. Start (5 km): the
participants started on the right lane of the motorway at a speed of 120 km/h during normal visibility and
100 km/h in fog (‘flying’ start). After this, the participants could choose their own speed, headway, lane, et
cetera. At the end of this scenario, the vehicle density increased. Also trucks and buses were simulated:
approximately 10% of the volume on the right lane. The traffic jam was still far ahead of the participants. Section 1 (3.5 km): at the beginning of this scenario, the participants with the Congestion Assistant received the
first congestion warning on the display. Otherwise, no information on the traffic jam at hand was given. The
vehicle density further increased as the traffic jam came closer. Section 2 (1.5 km): during this scenario, the
participants with the Congestion Assistant felt a counterforce on the gas pedal when approaching the tail
of the traffic jam at too high speed according to the system. The congestion warning was still displayed.
No information or assistance was provided in the runs without the Congestion Assistant. The vehicle density
further increased as participants were almost entering the traffic jam. Section 3 (3 km): this scenario consisted
of the traffic jam with vehicles on both lanes driving in stop-and-go mode. The Congestion Assistant took over
the longitudinal driving task from the participants and displayed information about the length of the traffic jam. Without the system, the participants had to drive in congestion themselves and received no jam information. Section 4 (2 km): at the beginning of this scenario, the participants were leaving the traffic jam.
144
C.J.G. van Driel et al. / Transportation Research Part F 10 (2007) 139–152
Participants driving with the Congestion Assistant had to take over control from the Stop & Go function. All
vehicles accelerated again and the vehicle density decreased.
2.4. Driving simulator
The experiment was conducted in TNO’s advanced driving simulator (Fig. 3). The participant was seated in a
BMW 318i mock-up that was placed on a motion base with six degrees of freedom. The mock-up was fitted with
original controls, such as pedals and steering wheel, and automatic transmission. The participant watched a large
radial screen (120 horizontally and 30 vertically) on which the road and traffic environment was projected by
three high-resolution (1280 · 1024) DLP projectors. Rear-view mirrors were simulated by means of separate
LCD displays. Also the sound of traffic in the environment and the sound of the simulator car were presented.
2.5. Participants
In total, 37 participants took part in the driving simulator experiment. These participants were selected
from the respondents who completed the user needs survey of Van Driel and Van Arem (2005). Based on their
answers, two groups of participants were formed: one group with a positive attitude towards congestion assistance (22 participants) and one group with a negative attitude (15 participants). All participants were between
25 and 59 years old (average 43, SD 10), had a driving licence for at least 5 years and drove regularly. They
were paid for their participation.
2.6. Procedure
Prior to the driving simulator experiment, the participants received a verbal outline of the research. Next, they
signed a form of informed consent, filled in several questionnaires and read information about the Congestion
Assistant. One experimental session took one morning or afternoon. Two participants participated in each
experimental session. While one participant was driving, the other participant completed one or more questionnaires and could rest. To get acquainted with the driving simulator and the Congestion Assistant, each participant started with a training run. After that, taking turns, the participants completed the four experimental runs.
They were told to drive as they would normally do in similar situations. Each experimental run lasted 15–20 min.
2.7. Data collection
The interaction between the driver and the Congestion Assistant was studied in terms of impacts on driving
behaviour and acceptance. Table 2 shows the dependent variables discussed in this paper. For driving behaviour, the variables were related to speed, acceleration, car-following and use of the Congestion Assistant. The
Fig. 3. The TNO driving simulator.
C.J.G. van Driel et al. / Transportation Research Part F 10 (2007) 139–152
145
Table 2
Dependent variables of driving behaviour and acceptance
Impacts on the driver
Dependent variables
Driving behaviour
Speed: speed (mean, SD)
Acceleration: acceleration (>0 m/s2; mean, maximum), deceleration (<0 m/s2; mean, maximum)
Car-following: time headway (mean, SD, cumulative), Time-To-Collision (TTC) (minimum)
Use of the Congestion Assistant (e.g., overruling the active gas pedal, reaction time after deactivation of Stop &
Go)
Acceptance
Acceptance: score on Van der Laan questionnaire
driving simulator stored these data with a frequency of 10 Hz. The acceptance questionnaire of new technological equipment developed by Van der Laan et al. (1997) was used to measure the acceptance of the
Congestion Assistant. The participants had to fill in checklists before driving and after each experimental
run with the Congestion Assistant: one checklist for the total system and a checklist for each function of
the system.
2.8. Statistical analysis
Most statistical analyses were performed by means of analysis of variance (ANOVA) for repeated measurements. For the analysis of driving behaviour, the within-subject factors were Visibility (2 levels), System (2
levels) and Section (4 levels). For the analysis of acceptance, these factors were Condition (3 levels), Experience (3 levels), Dimension (2 levels) and Function (3 levels). Furthermore, the between-subject factor Attitude
(2 levels) was considered in the analysis of driving behaviour. Because the motion base broke down halfway
the experiment, the factor Motion (2 levels) was included as a between-subject factor in this analysis as well.
Main effects and two-way interaction effects were studied. When p < .05, the results were considered to be statistically significant. Tukey HSD post hoc tests were used to find out which groups differed from each other.
The data concerning the minimum TTC were not restricted (e.g., bounded by a certain threshold), so that the
values ranged from almost 0 s to over 1000 s. Instead of ANOVA, the percentages of a minimum TTC below
4 s and 20 s were studied. Prior to the ANOVAs, Kolmogorov–Smirnov tests were performed to check
whether the dependent variables were normally distributed. The variables discussed in this paper were found
to have a normal distribution. The Kolmogorov–Smirnov test was used to detect differences in the shapes of
the cumulative distributions of time headway. This test was applied to the four conditions within each section,
resulting in six comparisons between two conditions. Therefore, the p-level of .05/6 = .008 was used, including
the Bonferroni correction. Biserial correlation coefficients (Field, 2000) were calculated to study the relation between the attitude towards congestion assistance in general and the acceptance of the Congestion
Assistant.
3. Results
In this section, the main impacts of the Congestion Assistant on driving behaviour – in particular speed,
acceleration, car-following and use of the system – and acceptance are discussed. For a more detailed discussion of the total experiment, see Van Driel and Van Arem (2006). This reference also includes the impacts of
the Congestion Assistant on mental workload and a detailed analysis of the acceptance results (e.g., the relation between acceptance and willingness to buy the Congestion Assistant).
3.1. Speed behaviour
The results concerning the mean speed revealed an interaction effect between System and Visibility
[F(1,33) = 11.56, p < .01] (Fig. 4). During normal visibility, the mean speed with the Congestion Assistant
was lower than without the system. However, the Congestion Assistant did not affect the mean speed in
fog. The mean speed during normal visibility was higher than in fog, regardless of driving with the system.
146
C.J.G. van Driel et al. / Transportation Research Part F 10 (2007) 139–152
Fig. 4. Mean speed as a function of Visibility and System.
Fig. 5. Mean speed as a function of System and Section.
Furthermore, an interaction effect between System and Section was found on the mean speed
[F(3,99) = 32.39, p < .001]. Fig. 5 shows that the mean speed with the Congestion Assistant was lower than
without the system when approaching the traffic jam (Section 2). In the other road sections, the Congestion
Assistant did not affect the mean speed.
The results concerning the standard deviation of speed revealed an interaction effect between System and
Section [F(3,99) = 10.44, p < .001]. The standard deviation of speed with the Congestion Assistant (16.6 km/h)
was larger than without the system (9.3 km/h) when approaching the traffic jam. This can be explained by the
fact that in this section the active gas pedal gradually reduced the speed (e.g., from 120 to 50 km/h), which
resulted in a larger standard deviation of the driver’s speed. The Congestion Assistant did not affect the standard deviation of speed in the other road sections.
3.2. Acceleration behaviour
The results concerning the mean acceleration revealed an interaction effect between System and Section
[F(3,99) = 13.78, p < .001]. The Congestion Assistant led to a lower mean acceleration in the traffic jam than
without this system (.62 versus .75 m/s2). So, the Congestion Assistant accelerated less hard in the traffic jam
than drivers without the system did. A similar result was found by the interaction effect between System and
C.J.G. van Driel et al. / Transportation Research Part F 10 (2007) 139–152
147
Section on the maximum acceleration [F(3,99) = 27.57, p < .001]. It appeared that the Congestion Assistant
speeded up with lower maximum accelerations in the traffic jam than drivers without the system did (2.57 versus 3.70 m/s2).
Furthermore, an interaction effect between System and Section was found on the mean deceleration
[F(3,99) = 21.78, p < .001] (Fig. 6). The Congestion Assistant led to a lower mean deceleration level in the traffic jam than without this system. So, the Congestion Assistant braked less hard in the traffic jam than drivers
without the system did. The results concerning the maximum deceleration revealed that similar maximum
decelerations were found in each section, regardless of driving with or without the Congestion Assistant. However, the main effect of System showed that the maximum deceleration level with the Congestion Assistant
( 2.50 m/s2) was lower than without this system ( 2.87 m/s2).
3.3. Car-following behaviour
An interaction effect between System and Section was found on the mean time headway [F(3,99) = 32.77,
p < .001]. Fig. 7 shows that when driving in the traffic jam and leaving the traffic jam (Sections 3 and 4), the
mean time headway with the Congestion Assistant was smaller than without the system. However, the
Fig. 6. Mean deceleration as a function of System and Section.
Fig. 7. Mean time headway as a function of System and Section.
148
C.J.G. van Driel et al. / Transportation Research Part F 10 (2007) 139–152
Fig. 8. Cumulative distributions of time headway before the traffic jam (a) and in the traffic jam (b) without the Congestion Assistant
(condition 1) versus with this system (condition 2) during normal visibility.
Congestion Assistant did not affect the mean time headway when driving (far) before the traffic jam (Sections 1
and 2).
The cumulative distributions of time headway showed differences between the time headways before and in
the traffic jam (p < .008). When approaching the traffic jam during normal visibility, about 55% of the time
headways were below 2 s when driving without the Congestion Assistant, whereas this was about 20% when
driving with the system (Fig. 8a). In fog, these percentages were about 25% when driving without the Congestion Assistant and about 20% when driving with the system. Evidently, the Congestion Assistant led to larger
time headways when approaching the traffic jam, especially during normal visibility. When driving in the traffic jam, about 25% of the time headways were below 2 s when driving without the Congestion Assistant,
whereas this was about 80% when driving with the system, regardless of the visibility conditions. So, the Congestion Assistant led to smaller time headways in the traffic jam (Fig. 8b).
The results concerning the standard deviation of time headway revealed an interaction effect between System and Section [F(3,99) = 17.09, p < .001]. When driving in the traffic jam, the standard deviation of time
headway with the Congestion Assistant (.47 s) was smaller than without the system (.87 s).
Furthermore, the Congestion Assistant affected the percentages of a minimum TTC below a certain threshold. When approaching the traffic jam, the percentages of a minimum TTC below 20 s with the Congestion
Assistant (39.2%) were lower than without this system (83.8%), indicating safe following situations with the
Congestion Assistant. When driving in the traffic jam, the percentages of a minimum TTC below 4 s with
the system (56.8%) were higher than without (31.1%). Generally, minimum TTCs below 4 s indicate potentially
dangerous following situations. However, it is not clear whether this also applies to systems that (partly) take
over the driving task, such as the Stop & Go function of the Congestion Assistant. For example, automating the
car-following task during congestion can eliminate possible human errors during this driving task. When leaving the traffic jam, the percentages of a minimum TTC below 20 s with the system (64.9%) were higher than
without (50.0%), indicating less safe following situations after having driven with the Congestion Assistant.
3.4. Use of the Congestion Assistant
All participants felt a counterforce of the active gas pedal of the Congestion Assistant at some point before
the traffic jam. Thus, no one was slowing down without triggering the active gas pedal. Generally, participants
experienced the first time of counterforce when they were about 500–700 m away from the traffic jam,
although this distance varied a lot between the participants. On average, the participants overruled the active
gas pedal two times when approaching the traffic jam. Seven participants did not overrule the pedal at all,
while one participant overruled it eight times (i.e., maximum).
Participants needed a reaction time to release or press the gas pedal when the Stop & Go of the Congestion
Assistant switched on or off. The activation of the Stop & Go function was introduced by a spoken message.
Five participants already released the gas pedal before the end of this message. Thus, these persons handed
over the control of their vehicle to the Stop & Go too early. More than half of the participants still had their
foot on the gas pedal when the Stop & Go took over the longitudinal driving task. Possibly, people wanted to
check the system’s operation first before releasing the pedal (i.e., trust in system). Also the deactivation of the
Stop & Go function was introduced by a spoken message. Seven participants already used the gas pedal before
C.J.G. van Driel et al. / Transportation Research Part F 10 (2007) 139–152
149
Fig. 9. Acceptance scores of the Congestion Assistant and its functions.
the end of this message. Thus, these persons took over the control of their vehicle from the Stop & Go fairly in
time. Around one third of the participants did not have their foot on the gas pedal when the Stop & Go
switched off. These people needed a reaction time (on top of the time between the spoken message and the
Stop & Go actually turning off 2 s) to press the pedal. More than 80% of the participants used the pedal
within 1 s. The mean reaction time of taking over from the Stop & Go was .6 s.
3.5. Acceptance
Generally, the participants stated that they appreciated the Congestion Assistant. They thought it could
help increase traffic safety and efficiency and decrease emissions. Fig. 9 shows the acceptance scores – split
up into a usefulness scale (u) and a satisfaction scale (s) – for the Congestion Assistant as a whole and for
the three functions separately. A distinction is made between before driving, after driving during normal visibility and after driving in fog.
It can be seen that the acceptance scores of the Congestion Assistant did not surpass the acceptance scores
of its functions. In particular, the warning function was highly valued, whereas the active gas pedal received
least acceptance. Participants thought that the Stop & Go was as useful as satisfying, while they regarded the
warning and the active gas pedal as more useful than satisfying [F(2,72) = 22.36, p < .001]. Besides, the participants were more negative about the Stop & Go before driving with the Congestion Assistant than after
driving with the system, regardless of visibility [F(4,144) = 6.90, p < .001]. Fig. 10 shows that gaining experience with the Congestion Assistant (i.e., before versus first run) led to a higher acceptance score of the Stop &
Go, while gaining more experience (i.e., first versus second run) did not affect the acceptance scores of the
three functions [F(4,144) = 7.27, p < .001].
Furthermore, the extent to which the attitude towards congestion assistance in general affected the acceptance of the Congestion Assistant was studied. Therefore, biserial correlation coefficients between the acceptance scores and the attitude towards congestion assistance – based on the participants’ answers to the user
needs survey – were calculated. It appeared that participants who were more positive about congestion assistance in advance were also more positive about the warning and Stop & Go functions of the Congestion Assistant (p < .05). However, this relationship mainly existed for the satisfaction scores before gaining experience
with the system. Apparently, the indicated user needs for congestion assistance particularly corresponded to
how satisfying one thought such a system would be.
4. Discussion
The main objective of this driving simulator experiment was to assess the impacts of the Congestion Assistant on the driver in terms of driving behaviour and acceptance. As can be seen below, the results confirmed
most of the earlier stated hypotheses.
When driving with the Congestion Assistant, the participants received a congestion warning when they
were approaching the traffic jam. It was expected that the participants were better prepared for the traffic
150
C.J.G. van Driel et al. / Transportation Research Part F 10 (2007) 139–152
Fig. 10. Acceptance scores as a function of Experience and Function.
situation ahead. This could be expressed by, for example, lower speeds and larger time headways. However,
the results did not show any impacts of the congestion warning on driving behaviour when the participants
were at a distance of 5–1.5 km from the traffic jam. This might be explained by the fact that they were still
relatively far away from the traffic jam. Remarks from some of the participants confirmed this. They thought
that a warning at 5 km from the traffic jam was ‘too early’. This is in line with the general practice of showing
speed limits on Variable Message Signs (VMS) above Dutch roads from 1.5 km before an accident until the
incident (Remeijn, 1998). Noecker, Karanasiou, Mitropoulos, and Strassberger (2005) discussed the ‘dangers’
of too early warnings (e.g., irritation, disbelief), assuming that a hazard warning at 1 km from the hazard as
being too early.
The active gas pedal and the congestion warning were both active when running into the traffic jam. However, it was assumed that particularly the active gas pedal affected the driving behaviour in this road section.
The active gas pedal produced a counterforce on the gas pedal when the participant was approaching the traffic jam at too high speed. Therefore, it was expected that the participants would better anticipate the downstream jam by earlier and smoother decelerations. The results showed a lower mean speed and a larger
minimum TTC with the Congestion Assistant when approaching the traffic jam. We also found a larger standard deviation of speed in this section when driving with the active gas pedal. This was an inherent part of the
operation of this function as it gradually reduced speed toward the jam (e.g., from 120 to 50 km/h). On
the contrary, without the active gas pedal, the participants continued driving at a high speed until noticing
the traffic jam, which resulted in smaller speed differences. The active gas pedal did not lead to a lower deceleration level. So, the participants braked equally hard regardless of driving with or without the active gas
pedal. All participants felt a counterforce on the gas pedal at some point. In general, they felt this counterforce
when being about 500–700 m away from the traffic jam. So, although the active gas pedal could warn the driver much earlier (namely from 1.5 km before the jam), most participants noticed the actual effects of the active
gas pedal later. Only few participants did not overrule the active gas pedal. Generally, the participants overruled the active gas pedal about two times.
In the traffic jam, the Stop & Go and the congestion information were both in operation. However, because
the participants could not overrule the Stop & Go, the effects on driving behaviour in this section were due to
the Stop & Go. By automating the longitudinal driving task, it was expected that the Stop & Go would perform the car-following task more efficiently. The results showed that the Stop & Go led to a lower mean and
maximum acceleration and to a lower mean deceleration level. So, the Stop & Go accelerated and braked less
hard than drivers without the system did. The acceleration results could indicate a slower response of the Stop
& Go to the other traffic. However, the time headway results seem to show the opposite. The Stop & Go operated at a time headway of 1 s. According to the results on mean time headway, the Stop & Go indeed followed
a lead vehicle closer (i.e., at smaller time headways) than the drivers ‘normally’ did. More time headways
C.J.G. van Driel et al. / Transportation Research Part F 10 (2007) 139–152
151
between 1 and 2 s were found, however, no differences were found between the following behaviour of the Stop
& Go and the participants with respect to the minimum time headway. The Stop & Go also led to a smaller
minimum TTC. Besides, the standard deviation of time headway with the Congestion Assistant was smaller
than without the system. These smaller differences in time headway due to the Stop & Go possibly indicate
a better following behaviour because of a more homogeneous headway distribution. After the traffic jam,
the mean time headway with the Congestion Assistant was smaller than without the system. Probably, the participants got used to driving at short time headways because the Stop & Go also displayed this behaviour in
the traffic jam. The (de)activation of the Stop & Go was introduced by a spoken message. Most participants
used the time between this message and the actual (de)activation to prepare for the Stop & Go actually turning
on or off. For example, more than half of the participants already got their foot on the gas pedal when the
Stop & Go switched off. The mean reaction time of taking over from the Stop & Go was about .6 s. This
is much faster than the 2 s found by Reichardt (1998). However, in his study the drivers were working in
the car (e.g., office on wheels) during automatic car driving, while in our study this was not the case.
The total acceptance of the system was rather high: an average score of 1.0 on a scale from 2 to +2. However, not all functions of the Congestion Assistant were equally appreciated. In general, systems that restrict
the driver’s behaviour are likely to be less accepted than non-restrictive, informative systems (Van der Laan
et al., 1997). However, this was not confirmed by our study. It appeared that the (informing) congestion warning as well as the (controlling) Stop & Go was highly accepted. Probably, drivers appreciated to be released by
the Stop & Go from the uncomfortable task of congestion driving. Research shows that in most cases drivers
were more positive about a driver support system after having gained experience with it (TRG, 2004). Therefore, it was expected that the acceptance of the Congestion Assistant was higher after one had driven with the
system. However, the data did not clearly support this expectation. Gaining experience (i.e., before versus first
run) led to a higher acceptance score of the Stop & Go, while gaining more experience (i.e., first versus second
run) did not affect the acceptance scores of the three functions. Furthermore, the results of the user needs survey indicated strong needs for help in reduced visibility situations (Van Driel & Van Arem, 2005). However,
the expectation that the Congestion Assistant would be accepted more in fog than during normal visibility was
not confirmed by the results. According to our expectations, it appeared that participants with a positive attitude towards congestion assistance – based on their survey answers – were more positive about the Congestion
Assistant than participants with a negative attitude. It can be concluded that the user needs survey revealed to
be a valid method for the indication of user needs for congestion assistance.
The Congestion Assistant was tested in specific congested traffic situations. Although all participants
encountered a traffic jam during their experimental runs, the appearance of this jam could differ between
the runs due to the surrounding traffic behaving differently (but within certain boundaries). More research
would be needed to study whether the results of this driving simulator experiment were sensitive to these
variations.
5. Conclusions
Generally, the participants stated that they appreciated the Congestion Assistant. They thought it could
help increase traffic safety and efficiency and decrease emissions. The results on driving behaviour also pointed
in that direction. For example, speed reductions caused by the active gas pedal could enhance traffic safety.
Furthermore, smaller time headways by the Stop & Go might enhance traffic efficiency, while lower acceleration and deceleration levels could indicate fewer emissions.
Not all functions of the Congestion Assistant were equally rated. The congestion warning and the Stop &
Go were accepted most, the active gas pedal least. To increase the acceptance of the Congestion Assistant, the
following suggestions for enhancement were made. Congestion warnings should not be given too early. For
example, the first congestion warning at 5 km from the traffic jam might be too soon. The active gas pedal
should operate more as a final warning. For example, it could provide haptic feedback only when the driver
is very near to the tail of the traffic jam (e.g., 500 m). The Stop & Go should be overrulable. For example, the
driver could overrule this function by using the gas or brake pedal.
Summarizing, the Congestion Assistant showed promising improvements in traffic safety when approaching a traffic jam due to the active gas pedal, although the participants did not express great appreciation of this
152
C.J.G. van Driel et al. / Transportation Research Part F 10 (2007) 139–152
function. Furthermore, positive effects on traffic safety and efficiency in a traffic jam can be expected by the
Congestion Assistant due to the Stop & Go, a function which was highly appreciated by the participants.
The next step of this project will study these potential impacts on the traffic flow in more detail by means
of a microscopic traffic simulation study. The detailed information provided by this driving simulator experiment will be used as input for this study. For example, the effects of varying equipment rates and parameters
of the Congestion Assistant are topics of interest.
Acknowledgements
This project is part of the research program of knowledge centre Applications of Integrated Driver Assistance (AIDA) that has been realised by TNO and the University of Twente in the Netherlands. The project
also falls under the Dutch knowledge investment program Transitions towards Sustainable Mobility (TRANSUMO), project Intelligent Vehicles.
References
Benz, T., Christen, F., Lerner, G., & Vollmer, D. (2003). Traffic Effects of Driver Assistance Systems – The Approach within INVENT.
Paper presented at the 10th World Congress on ITS, Madrid, Spain.
Delphi. (2005). Delphi Forewarn Smart Cruise Control with Headway Alert & Stop-and-Go, Available from: http://www.delphi.com/pdf/
ppd/safesec/for_stopgo.pdf, consulted: January 2006.
European Commission. (2001). White Paper; European transport policy for 2010: time to decide, Brussels, Belgium.
European Commission. (2002). Final Report of the eSafety Working Group on Road Safety, Brussels, Belgium.
Field, A. (2000). Discovering Statistics Using SPSS for Windows. London, United Kingdom: SAGE Publications Ltd.
Hogema, J. H., & Rook, A. M. (2004). Intelligent speed adaptation: the effects of an active gas pedal on driver behaviour and acceptance,
Report no. TM-04-D011, TNO Human Factors, Soesterberg, The Netherlands.
Hogema, J. H., & Van der Horst, A. R. A. (1994). Driving behaviour in fog: analysis of inductive loop data, Report no. TM 1994 C-6, TNO
Human Factors, Soesterberg, The Netherlands.
Hummel, M., Siedersberger, K., Zavrel, M., Weilkes, M., Bürkle, L., Sauerbrey, J., et al. (2003). Traffic Congestion Assistance within the
Low Speed Segment. Paper presented at the 10th World Congress on ITS, Madrid, Spain.
Johansson, E., Carsten, O., Janssen, W., Jamson, S., Jamson, H., Merat, N., et al. (2005). HASTE; Validation of the HASTE protocol
specification, Deliverable 3.
Krautter, W., Mackamul, H., Manga, B., & Manstetten, D. (2004). Traffic scenarios in driving simulation: implementation and
application, Advances in Transportation Studies, 2004 Special Issue, pp. 9–18.
Malone, K. M., & Van Arem, B. (2004). Traffic effects of inter-vehicle communication applications in CarTALK 2000. Paper presented at
the 11th World Congress on ITS, Nagoya, Japan.
Marsden, G., Brackstone, M., & McDonald, M. (2001). Assessment of the Stop and Go function using real driving behaviour. Paper
presented at the 2001 International Conferenceon Advanced Driver Assistance Systems, Birmingham, United Kingdom.
Michon, J. (1985). A critical view of driver behavior models: what do we know, what should we do? In L. Evans & R. Schwing (Eds.),
Human behavior and traffic safety. New York, USA: Plenum Press.
Ministry of Transport, Public Works and Water Management. (2004). Nota Mobiliteit; Naar een betrouwbare en voorspelbare
bereikbaarheid, The Hague, The Netherlands (in Dutch).
Monsell, S. (2003). Task switching. Trends in Cognitive Sciences, 7(3), 134–140.
Noecker, G., Karanasiou, I., Mitropoulos, G., & Strassberger, M. (2005). WILLWARN; Warning strategies, Deliverable D22.34/I.
Regan, M., Mitsopoulos, E., Haworth, N., & Young, K. (2002). Acceptability of in-vehicle intelligent transport systems to Victorian car
drivers, Report no. 02/02, Monash University Accident Research Centre, Melbourne, Australia.
Reichardt, D. M. (1998). Using automated assistance systems – Putting the driver into focus. Paper presented at the IEEE International
Conference on Intelligent Vehicles (IV), Stuttgart, Germany.
Remeijn, H. (1998). Levels of quality of service for VMS systems. Paper presented at the 5th ITS World Congress, Seoul, South Korea.
Stanton, N. A., & Young, M. S. (1998). Vehicle automation and driving performance. Ergonomics, 41(7), 1014–1028.
TRG. (2004). STARDUST; Final report, Deliverable 15.
Van der Laan, J. D., Heino, A., & De Waard, D. (1997). A simple procedure for the assessment of acceptance of advanced transport
telematics. Transportation Research Part C, 5(1), 1–10.
Van Driel, C. J. G., & Van Arem, B. (2005). Integrated driver assistance from the driver’s perspective; Results from a user needs survey.
Enschede, The Netherlands: University of Twente, Centre for Transport Studies.
Van Driel, C. J. G., & Van Arem, B. (2006). Impacts of a Congestion Assistant on driving behaviour, workload and acceptance; Results from
a driving simulator study. Enschede, The Netherlands: University of Twente, Centre for Transport Studies.
Venhovens, P., Naab, K., & Adiprasito, B. (2000). Stop and Go Cruise Control. International Journal of Automotive Technology, 1(2),
61–69.