Safety Science 47 (2009) 620–625

Contents lists available at ScienceDirect

Safety Science
journal homepage: www.elsevier.com/locate/ssci

Effects of time-gap settings of adaptive cruise control (ACC) on driving

performance and subjective acceptance in a bus driving simulator
Tsang-Wei Lin a, Sheue-Ling Hwang a,*, Paul A. Green b
a
Department of Industrial Engineering and Engineering Management, National Tsing Hua University, 101 Guangfu Road Section 2, Hsinchu 300, Taiwan
b
University of Michigan Transportation Research Institute, Human Factors Division, 2901 Baxter Road, Ann Arbor, MI 48109-2150, USA

a r t i c l e i n f o a b s t r a c t

Article history: A study was conducted to investigate the effects of time-gap settings and contents of secondary tasks on a
Received 24 January 2007 fix-based bus driving simulator on drivers’ performance while reclaiming control from ACC in a car-fol-
Received in revised form 7 August 2008 lowing scenario of emergency brake by the lead vehicle. Thirty professional bus drivers drove on the sim-
Accepted 7 August 2008
ulator with the scenario of highway traffic flow under 12 random time-gap settings: from 0.64 s to 2.40 s
with the interval of 0.16 s. As for the effects of secondary tasks, subjects were evenly divided into three
conditions: no secondary task interference, simple secondary task, and complex task. The results demon-
Keywords:
strated that different safety demarcations of time-gaps on subjective acceptance and driving performance
Adaptive cruise control (ACC)
Time-gap settings
can be found out. The integrated demarcations separated time-gaps into divisions that represented dif-
Safety ferent levels of danger. It revealed that the safer time-gaps for different situations were: longer than
Bus 1.60 s for none-secondary task distraction and longer than 2.08 s for being continuously distracted by
secondary tasks. The demand for simple tasks is relatively high, so a larger time-gap is needed for the
driver to remain safe. This research has implications for the time-gap selection of ACC and effects of sec-
ondary task distraction on buses. A next logical step will focus on determining time-gaps for lead vehicles
on curves or slopes, when multiple vehicles are present ahead, and modeling driver behavior and perfor-
mance with ACC for cars, buses, and other types of vehicles.
Ó 2008 Elsevier Ltd. All rights reserved.

1. Introduction 1.1. How ACC affects driving performance

Driving is a dynamic control process that at times can be over- ACC systems can have positive or negative effects on traffic flow
whelming. Nowadays, automated driving assist systems have been and safety. For traffic flow, some drivers may choose longer time-
developed and implemented to make that task less overwhelming. gaps to reduce the number of mismatches in drivers’ expectations
In general, automated driving assistance systems are expected to of when and how the ACC decelerates (Zheng and McDonald,
improve system and operator safety, efficiency, and comfort 2005), which will reduce highway capacity. However, overall,
(Ward, 2000). One such driving automation system, adaptive ACC leads to more gradual acceleration and deceleration of traffic
cruise control (ACC), is now standard equipment for some upscale streams, improving traffic flow.
passenger cars (e.g., the BMW 500/700 series after 2003 and Audi Because an ACC system maintains a gap to the vehicle in front
A6 after 2005). that is adjusted for the driving conditions, providing ACC systems
ACC is an enhancement of conventional cruise control systems, should make driving safer. Therefore, the driver should have more
which sets a fixed driving speed regardless of traffic. With ACC, the resources available to attend to other tasks, such as looking at
driver is not only a controller, but a supervisor for some critical route signs and traffic signals, or potentially overloading in-vehicle
driving tasks. ACC can maintain a driver-selected ‘‘time-gap” tasks such as conversing on a cell phone and monitoring informa-
between the front bumper of the subject’s vehicle and the rear tion systems. Further, while using ACC, previous research indicates
end of the lead vehicle by measuring that distance with radar or that the maximum speed (Bjørkli and Jenssen, 2003; Törnros et al.,
lidar and the speed of the subject’s vehicle (SAE, 2003). Further- 2002), mean speed and headway variance (Ward, 2000) are less, a
more, in addition to controlling the engine and power train, ACC safer condition. Moreover, the speed variance (Bjørkli and Jenssen,
also provides some automatic operation of the brakes (SAE, 2003). 2003; Suzuki and Nakatsuji, 2003; Törnros et al., 2002; Ward,
2000) and headway variance (Fancher et al., 1998) are also less,
* Corresponding author. Tel.: +886 35742694. which reduces the likelihood of crashes. Also, when ACC is pro-
E-mail address: [email protected] (S.-L. Hwang). vided, drivers pay more attention to lateral control than when

0925-7535/$ - see front matter Ó 2008 Elsevier Ltd. All rights reserved.
doi:10.1016/j.ssci.2008.08.004
 T.-W. Lin et al. / Safety Science 47 (2009) 620–625 621

ACC is not provided (Ohno, 2001), which should reduce lane sion), the minimum gap during an encounter, and the number of
departures. crashes, and (3) how time-gaps are affected by the complexity of
In contrast, drivers may glance ‘‘off-road” more frequently and concurrent secondary tasks.
longer when using ACC (Thompson et al., 2006), which decreases
safety. Studies have found ACC brought about larger brake and 2. Method
acceleration forces (Bjørkli and Jenssen, 2003; Törnros et al.,
2002), especially at higher speeds (Hoedemaeker and Brookhuis, 2.1. Participants
1998), potentially leading to a loss of control. Finally, ACC could also
lead to shorter minimum time-gap (Hoedemaeker and Brookhuis, Thirty professional bus drivers (all male, between 26 and 53
1998; Törnros et al., 2002). years of age; 15 city bus drivers and 15 expressway bus drivers)
were paid NT$2000 ($60) for participating in this experiment. All
1.2. Research on time-gaps participants had a valid professional bus driver’s license for at least
three years and no crashes recorded for their current employer. All
Contemporary ACC systems are designed primarily for use in were employed by motor transport companies and drove at least
smooth flowing, light-to-heavy traffic at highway speeds, not stop 40 h per week.
and go traffic or emergency braking situations. Accordingly, when
an unexpected condition occurs that may require emergency brak- 2.2. Apparatus
ing, the driver will need to intervene and react. Normally, how
much time do drivers need to avoid accidents? At present, there Data was collected in custom-made, fixed base, bus driv-
is research on ACC operating characteristics for cars, but almost ing simulator at Chung Hua University, Taiwan. The intercity
no research for buses. To provide context, Table 1 lists some rec- bus simulated was for 10 m (length)  3 m (width)  4 m
ommended time-gap ranges for cars. Some characteristics are con- (height).
sidered for these ranges, such as drivers’ sex and age (Fancher The simulator used consisted of two subsystems. The sensing
et al., 1998), and roads (Törnros et al., 2002). subsystem consisted of a bus cab with steering wheel, accelerator,
brake, and dashboard with speedometers connected to the main
1.3. Importance computer (Intel Pentium IV 3.0G). Speed was shown on a head-
up display (HUD) located in front of the driver, at 15° below line
Buses differ in many ways from cars. The driver eye height for of sight of the driver who was 180 cm in height. The digits were
buses (280 cm) is greater than for cars (120 cm), allowing drivers 15 cm (height)  12 cm (width). The eye to screen distance was
to see farther down the road, which should lead to shorter de- 1.5 m. In simulators, drivers often underestimate their speed, and
sired time-gaps, but buses have much larger stopping distances this HUD was intended to make that less likely (Godley et al.,
(80 m from 100 km/h) than cars (40 m), which should increase 2002; Törnros, 1998).
desired time-gaps. Buses have a much lower horsepower to The interface between the simulator and the computer was a NI
weight ratio (hp/lb) than cars (0.009–0.013 vs. 0.045–0.16), so PCI-6034E analog card, and the software, Virtools Dev 3.0, gener-
they accelerate more slowly, making precise control of a time- ated the 3-D models (of cars, trees, roads, etc.), and controlled traf-
gap more difficult. Furthermore, buses have a lower static stabil- fic flow and car motion. The graphics subsystem controlled the
ity factor (SSF) than passenger cars, with lower rollover resistance information flow between the main computer (server) and two cli-
ratings (1–1.2 vs. 1.3–1.5), so swerving is less of an option than ent computers (Pentium IV 2.0G), each of which served as an image
braking. Finally, because bus occupants are much more likely to generator for side projectors. (The main computer controlled the
be passengers rather than drivers and not in control of the vehi- central projector.) Communication between the server and client
cle, more sedate acceleration is desired. In summary, there are so was controlled by Visual C++ code from the software development
many differences between the two vehicle types, the recom- kit (SDK) of Virtools Dev 3.0. The total horizontal field of view
mended ACC settings for buses may be different from cars and (FOV) of the three projectors was 160°.
should be determined. As typical for buses of this type, the simulated bus had an auto-
matic transmission and was capable of 0.3 g maximum accelera-
1.4. Objectives of this research tion and 0.51 g maximum braking. The operating characteristics
and user interface to the ACC conformed with SAE J2399 standard
In this paper, three major issues relating to bus ACC systems (SAE, 2003). The maximum deceleration provided by the system
were addressed: (1) what ACC time-gaps (between 0.64 s and was 0.2 g. While the ACC system was active, the official ISO symbol
2.40 s) are preferred on expressway driving, (2) what time-gaps (ISO 7000-2580) was shown on the windscreen in front of the dri-
maximize safety, as assessed by statistics of the gap when braking ver. The symbol would disappear if the driver controlled the vehi-
begins, brake pedal movement time (from contact to 50% depres- cle manually.

Table 1
Previous results about recommended time-gaps on ACC (for cars)
Measures Number of Exposure duration Environment Max ACC Recommended time-gaps Source
subjects (Per subject) decel. level
Subjective preference 24 30 km Driving simulator  1.50–2.49 s (motorway) Törnros et al. (2002)
1.66–3.21 s (rural road)
2
Number of expectation None  Computational simulation 3 m/s 2 s or more Zheng and McDonald (2005)
mismatches
Headway time margin 108 519 km Equipped car 0.686 m/s2 1.1 s (young) Fancher et al. (1998)
1.5 s (middle-aged)
2.1 s (older)
Driver’s preference 38 15 min Driving simulator  No preference Hoedemaeker and Brookhuis (1998)
Subjective selection 9 600 km Equipped car  1.1–1.8 s Reichart et al., (1996)
622 T.-W. Lin et al. / Safety Science 47 (2009) 620–625

2.3. Tasks, experimental design, and procedure experiment involved decisions about the distance or time to a lead
vehicle, it was important to ensure their gap estimates were accu-
A two-way (12 time-gaps, within-subjects; 3 secondary tasks, rate. In total, the experiment lasted 80–100 min for each
between-subjects) factorial design was used in this study. Time- participant.
gaps were from 0.64 s to 2.40 s in 0.16 s increment (0.64 s, 0.80 s,
0.96 s,. . .,2.24 s, 2.40 s), which were selected based on consider- 2.4. Data collection and analysis
ation of prior ACC studies involving passenger cars. Time-gaps were
presented in random order. The primary task in this study was driv- Only part of each test session was of interest, namely from
ing on a straight and level three-lane highway (3.5 m wide/lane) when the lead vehicle began emergency braking until the lead
with a shoulder (1 m wide). Participants were asked to drive in vehicle reaccelerated and left. The gap to the lead vehicle, vehicle
the middle lane without overtaking, changing lanes, or exceeding speed, and the position of acceleration and brake pedal were all
the speed limit (100 km/h) while following the lead vehicle. Sec- collected at 10 Hz. Four statistics during the interval were com-
ondary tasks were separated into three levels: none, simple, and puted, (1) the gap when braking began (where they were when
complex tasks. The secondary task was the addition of two num- braking began), (2) brake pedal movement time (time from brake
bers with two-digits each (less than 50), such as 18 + 36, shown pedal contact until 50% pedal travel-how quickly they responded),
in the LCD monitor at the right-hand side of the driver. The maxi- (3) minimum gap (how close they were to the lead vehicle at any
mum response time was 8 s and the inter-trial interval was 3–5 s. time), and (4) the number of crashes (how many negative out-
A tone would occur as a cue just before the question appeared. In comes occurred). Also, driver’s subjective acceptance (1 = not
the simple secondary task, participants had to calculate the sum acceptable; 0 = acceptable) for each time-gap setting was collected
and answer verbally. For the complex task, participants needed to after each braking event.
calculate, answer, and then to press the button on the LCD monitor ANOVA was computed for three statistics, the gap when braking
(touch panel) to confirm. In other words, the complex task had one began, brake pedal movement time, and minimum gap, to examine
more step than the simple one, the confirmation. main effects and interactions. The non-parametric Cochran Q test
This experiment was divided into 12 blocks, one for each differ- was used to analyze the subjective acceptance and the number of
ent time-gap. Every block consisted of four trials, so there were 48 crashes. Then, the two-stage clustering (Punj and Stewart, 1983)
trials in all. Emergency braking of the lead vehicle would occur in was applied to discriminate subsets for 12 time-gap settings, and
one of the four trials in each block, where the lead vehicle would the LSD test was used to find out the difference among three sec-
unexpectedly decelerate at 0.82 g from 100 km/h to 20 km/h (but ondary tasks.
only if the subject vehicle reached 100 km/h). After the lead vehicle Ward’s hierarchical grouping method (Ward, 1963) was used to
reached the speed of 20 km/h, it would remain the speed for 5 s form mutually exclusive subsets (by specified characteristics) for
and reaccelerate to drive away. For the other three trials, the lead the first stage. The procedure was to unite n observations to n1
vehicle would only adjust the speed between 100 km/h and 80 km/ clusters repeatedly until only one group remained. At each round,
h with the deceleration of 0.2 g, again, remaining at 80 km/h for the distance between cluster centers could be found. The longer
five seconds before the lead vehicle reaccelerated and drove away. the distance was, the more distinct clusters would be. After that,
The unexpected emergency braking was beyond what the ACC time-gaps could be classified by K-means clustering (second stage),
could handle, so the driver had to reclaim control to avoid a crash. whose purpose was to maximize the within-cluster homogeneity
The trial on which emergency braking occurred varied from block and between-cluster separation with four main steps (Cox, 2005).
to block, as shown in Fig. 1. In this experiment, for two-stage clustering, 12 time-gaps might
To begin, participants were briefed about the function of the be taken as ‘‘observations” that should be classified, and 30 partic-
simulator. Next, the primary driving and secondary tasks were ex- ipants would be treated as ‘‘characteristics” for observations.
plained and subjects were told they could quit whenever they felt
dizzy or uncomfortable. Next, subjects practiced driving for 20– 3. Results
30 min. After that, subjects drove the 12 test blocks. At the end
of the experiment, each subject completed a test in which they 3.1. What time-gaps did subjects accept?
judged the distance to a lead vehicle (on the 2D projection, pre-
sented at 10 m, 20 m, and 40 m). At 100 km/h, these distances cor- There were significant differences on subjective acceptance for
respond to time-gaps of 0.36 s, 0.72 s, and 1.44 s. Since the time-gaps (Cochran’s Q = 220. 281, p < 0.001). The numbers were

Fig. 1. The test trials illustration.

 T.-W. Lin et al. / Safety Science 47 (2009) 620–625 623

shown in Fig. 2 and two subsets (time-gaps below 1.44 s and

above 1.60 s) were recommended by the clustering procedure,
which meant that time-gaps greater than 1.60 s were relatively
more acceptable than time-gaps below 1.44 s. When the time-
gap was increased from 1.44 s to 1.60 s, the proportion of accep-
tance rose to about 50%, that is, almost half of participating bus
drivers could accept 1.60 s as the current time-gap. If longer
time-gaps were selected, more than half of subjects could accept
them. How drivers’ subjective acceptance would affect their risk
avoidance strategies and behavior was analyzed in the following
sections.
Probability functions for decisions can often be determined
using logistic regression with two parameters, one of which repre-
sents the spread of the distribution and the second that described
where the yes/no transaction occurs. For Eq. (1), the percent con-
cordant was 85% and the p-value for the Hosmer and Lemeshow
test was 0.949. Eq. (1) presents the function for the subjective
acceptance and the probability a time-gap is unacceptable. An
equation from acceptance for any time-gap can be estimated, not
just those tested

e9:0415:222x
Pr ðCurrent time-gap is unacceptableÞ ¼ ;
1 þ e9:0415:222x
where x ¼ time-gaps ð0:64 s to 2:40 sÞ ð1Þ
From this model, time-gaps below 1.44 s and above 1.60 s
would cause the probability that current time-gap is unacceptable
to be greater than 0.82 and lower than 0.67.

3.2. Driver performance

The statistics of gap when braking began, brake pedal move-

ment time, minimum gap, and number of crashes were analyzed
in this section with ANOVA (Cochran’s Q test) and clustering. As
shown in Table 2, all statistics were significantly affected by
time-gaps. Only the interaction for the minimum gap was signifi-
cant. For the situations without interaction, Ward’s method indi-
cated there were two time-gaps subsets, with K-means clustering
showing the border was between 1.44 s and 1.60 s (Fig. 2a, c, and
e). As to the result in Fig. 2b, the line had a steadily continuous in-
crease with the difference of 4 m between each time-gap. It was
not obvious where the border was.
The simple main effects of the two factors were analyzed
through two-stage clustering (for time-gaps) and LSD method
(for secondary tasks). As shown in Fig. 2d, without secondary tasks
(the solid line in Fig. 2d), the border was also not obvious. The
slope was stable. However, for simple and complex secondary
Fig. 2. Effect of time-gap on (a) subjective acceptance (b) gap when braking began,
tasks (the broken line in Fig. 2d), the border could be 2.08 s, which (c) brake pedal movement time, (d) minimum gap, and (e) number of crashes.
corresponded to an increase in the minimum gap by 7.40 m (three
quarters of bus length). When being distracted by secondary tasks,
longer time-gaps were recommended.
The LSD test indicated that the gap when braking began with no e5:7454:126x
secondary task was significantly longer than with the simple and Pr ðCrashesÞ ¼ ;
1 þ e5:7454:126x
complex tasks. When comparing with no secondary task, simple where x ¼ time-gaps ð0:64 s to 2:40 sÞ ð2Þ
and complex secondary tasks slowed down drivers’ responses
and shortened the gap when braking began by 4.14 m and In this model, time-gaps below 1.44 s and above 1.60 s
6.57 m respectively. Further, the simple main effects of secondary would cause the crash mean to be greater than 0.45 and lower
tasks on minimum gap showed that the significance between with/ than 0.3.
without secondary task conditions occurred as long as time-gaps
were between 1.12 s and 1.92 s. Compared with no secondary task, 3.3. Visual test for simulation environment
minimum gaps with simple and complex secondary tasks were
shorter by 5.25 m and 7.14 m. Distances (gaps) estimated for 10 m, 20 m, and 40 m were
To take the analysis of crashes one step further, logistic regres- 9.5 m, 19.9 m, and 37.5 m, respectively, with a mean standard
sion was used to model the relationship between crash probability deviation of 0.8, 1.3, and 2.2 (8.4%, 6.7%, 6.0% of mean), so esti-
and time-gaps Eq. (2). The percent concordant was 83% and the mates of distance in the driving simulator were both accurate
p-value for the Hosmer and Lemeshow test was 0.164 and precise, supporting the use of the driving simulator.
624 T.-W. Lin et al. / Safety Science 47 (2009) 620–625

Table 2 belonged to the definition of ‘‘Visual read (symbol),” so the VACP

ANOVA results for driver performance variables and Cochran’s Q test result for demand could be rated with the score of 5.9 (see Yee et al.
number of crashes
(2007)). By the same way, the total demands for simple and com-
Statistics Source F d.f. p plex task in this study could be determined, which were 14.9 and
Gap when braking began Time-gaps (T) 316.611 11 <0.001* 21.1, with VACP values of (5.9, 1.0, 7.0, 1.0) and (9.9, 1.0, 7.0, 3.2),
TS 1.076 22 0.372 respectively. Also, the demands of other in-vehicle subtasks (e.g.
Error (T) 297
dial phone, eat food, hold a cigar, etc.) can be quantified using these
Secondary tasks (S) 12.352 2 <0.001*
Error (S) 27 ratings.
Brake pedal movement time Time-gaps (T) 27.344 11 <0.001* The consequences of a delayed response could be improved
TS 0.697 22 0.842 by selecting time-gaps of 2.08 s or longer, as larger gaps give
Error (T) 297
the driver more time to assess the situation and take evasive
Secondary tasks (S) 0.112 2 0.895
Error (S) 27
action. Stanton et al. also suggested that ACC should provide
Minimum gap Time-gaps (T) 76.423 11 <0.001* drivers with enough time to reclaim control from the auto-
TS 3.874 22 <0.001* mated system (Stanton et al., 1997). Long time-gaps, however,
Error (T) 297 could lead to delayed driver responses because the bus drivers
Secondary tasks (S) 10.206 2 0.001*
thought the current time-gap was not so urgent that they
Error (S) 27
could react later (Lin et al., 2008). Although the delayed re-
Statistics Source Cochran’s Q p sponse would not lead to danger because of the long time-
The number of crashes Time-gaps (T) 198.634 <0.001* gap, it could also bring about discomfort resulting rapid braking
Note: *statistical significant. or steering.
In most cases interactions were not statistically significant,
except the minimum gap. This phenomenon might be due to the
4. Discussion design of experiment and number of subjects in each experimental
condition. There were only 10 participants for each condition (30
4.1. Time-gaps participants evenly distributed into three secondary task condi-
tions). Increasing the number of participants or replicates in this
For most measures, the demarcation between riskier and safer experiment could lead to interactions that are statistically
conditions was for time-gaps between 1.44 s and 1.60 s, so time- significant.
gaps above 1.60 s are recommended for buses for all secondary task
conditions. However, the result is not consistent with some previ-
4.2. Limitations
ous research for cars (in Table 1). Hoedemaeker and Brookhuis
(1998) indicated that drivers had no significant preference for
This research does have some limitations. The traffic density of
time-gaps based on rating scale mental effort (RSME). Time-gaps
the simulator is about 1080 vehicles per hour per lane with the
in Zheng and McDonald’s research, 2005 were recommended to
speed of 110 km/h, which corresponds to Level of Service A on a
be more than 2 s. The difference could be due to that (1) subjects
freeway, the conditions under which ACC is most likely to be used.
were professional bus drivers and the environment was a bus driv-
When the traffic density is greater, the speed will be lower and the
ing simulation in this research, and (2) time-gaps in this research
required time-gap will change as well. Also, this experiment did
were from 0.64 s to 2.40 s, which covered wider time-gap ranges
not involve any task other than driving on a straight road and
than previous ones.
maneuvers other than braking. When driving on curves, depending
The experimental result showed that time-gaps above 1.60 s
on the radius and speed, some ACC systems lose and then re-ac-
were more acceptable, and more likely to prevent collisions, but
quire lead vehicles or may interpret vehicles in other lanes as being
the border would increase to 2.08 s if drivers were distracted by
lead vehicles, especially if both the target and subject vehicles are
secondary tasks. There was a huge time-gap difference (0.48 s,
close to the same lane boundary. This can cause the ACC system to
13.2 m) between these two situations, which could be taken as
slow the vehicle suddenly and severely. The extent to which this
the time for drivers to decouple from the secondary task. Fortu-
occurs varies with headway, traffic, and curvature, and not
nately, the negative effect of secondary tasks could be eliminated
necessarily in a consistent manner with increasing headway.
by longer time-gaps, as was the finding of previous research (Lin
Furthermore, preferred ACC settings should be different when
et al., 2008). It was also indicated that the driver would keep short-
maneuvering (changing lanes, cut-ins) as maneuvers often lead
er minimum gaps when driving with ACC (Hoedemaeker and
to the sudden appearance of a lead vehicle at a short range, which
Brookhuis, 1998; Törnros et al., 2002), but time-gaps greater than
leads to braking.
2.08 s were desired when distracted by a secondary task. In this re-
search, to reclaim control was a highly visual-manual demanding
task in which the driver observes the lead vehicle and perceives 5. Conclusions
if the emergency braking occurred. The visual, auditory, and cogni-
tive loads would lead to longer reaction time (Lee et al., 2001; Rich- The safety implications of various ACC time-gap settings were
ard et al., 2002). That is why the designed secondary tasks had such examined using five driving performance statistics. For all five,
a significant effect. the recommended time-gaps for ‘‘car following” on an expressway
Comparing with other studies about the effect of secondary at 100 km/h with light traffic (Level of Service A) were: (1) longer
tasks, one needs to quantify the demands of these tasks. Yee than 1.60 s for no secondary task distraction and (2) longer than
et al. rated the visual (V), auditory (A), cognitive (C), and psycho- 2.08 s for being continuously distracted by secondary tasks, such
motor (P) demands of 68 subtasks performed while driving. Rat- as monitoring in-vehicle information systems or speech interac-
ings were from the US Army IMPRINT modeling tool (0-to-7 tion. Even though drivers often misestimate distances and speeds
scale, with 7 being the most demanding) (Yee et al., 2007). For in simulators, the judgment task indicates the distance estimates
all four scales, subtasks could be rated according to their contents. were precise and accurate. A next logical step will focus on
For instance, the visual demand of the secondary task in this determining time-gaps for lead vehicles on curves or slopes, when
research (looking at the LCD monitor and read the numbers) multiple vehicles are present ahead, and modeling driver behavior
 T.-W. Lin et al. / Safety Science 47 (2009) 620–625 625

and performance with ACC. This should be done for cars, buses, and Punj, G., Stewart, D.W., 1983. Cluster analysis in marketing research: review
and suggestions for application. Journal of Marketing Research 20 (2), 134–
other types of vehicles as well.
148.
Reichart, G., Haller, R., Naab, K., 1996. Driver assistance: BMW solutions for the
Acknowledgement future of individual modality. In: Paper Presented at the Third Annual World
Congress on Intelligent Transport Systems, Orlando, FL, USA.
Richard, C.M., Wright, R.D., Ee, C., Prime, S.L., Shimizu, Y., Vavrik, J., 2002. Effect of a
This research was funded by the National Science Council (NSC) concurrent auditory task on visual search performance in a driving-related
of Taiwan (Contract No. NSC93-2218-E-216-013-). The authors image-flicker task. Human Factors 44 (1), 108–119.
would like to thank the Department of Transportation Technology SAE, 2003. Adaptive Cruise Control (ACC) Operating Characteristics and User
Interface: SAE International.
and Logistic Management, Chung Hua University, for providing the Stanton, N.A., Young, M., McCaulder, B., 1997. Drive-by wire: the case of driver
bus driving simulator. workload and reclaiming control with adaptive cruise control. Safety Science 27
(2–3), 149–159.
Suzuki, H., Nakatsuji, T., 2003. Effect of adaptive cruise control (ACC) on traffic
References throughput: numerical example on actual freeway corridor. JSAE Review 24 (4),
403–410.
Bjørkli, C.A., Jenssen, G.D., 2003. Adaptive cruise control (ACC) and driver Törnros, J., 1998. Driving behaviour in a real and a simulated road tunnel – a
performance: effects on objective and subjective measures. In: Paper validation study. Accident Analysis and Prevention 30 (4), 497–503.
Presented at the 10th World Congress on Intelligent Transportation Systems, Törnros, J., Nilsson, L., Östlund, J., Kircher, A., 2002. Effects of ACC on driver
Madrid, Spain, 16–20 November. behaviour, workload and acceptance in relation to minimum time headway. In:
Cox, T.F., 2005. An Introduction to Multivariate Data Analysis. Hodder Education, Paper presented at the Ninth World Congress on Intelligent Transportation
New York. Systems, Chicago, Illinois, 14–17 October.
Fancher, P., Ervin, R., Sayer, J., Hagan, M., Bogard, S., Bareket, Z., Mefford, M., Haugen, Thompson, L.K., Tönnis, M., Lange, C., Bubb, H., Klinker, G., 2006. Effect of active
J., 1998. Intelligent cruise control field operational test (Technical report DOT cruise control design on glance behaviour and driving performance. In: Paper
HS 808 849). National Highway Traffic Safety Administration, Washington, DC. presented at the 16th World Congress on Ergonomics, Maastricht the
Godley, S.T., Triggs, T.J., Fildes, B.N., 2002. Driving simulator validation for speed Netherlands, 10–14 July.
research. Accident Analysis and Prevention 34 (5), 589–600. Ward, J.H., 1963. Hierarchical grouping to optimize an objective function. American
Hoedemaeker, M., Brookhuis, K.A., 1998. Behavioural adaptation to driving with an Statistical Association Journal 58 (301), 236–244.
adaptive cruise control (ACC). Transportation Research Part F 1 (2), 95–106. Ward, N.J., 2000. Automation of task processes: an example of intelligent
Lee, J.D., Caven, B., Haake, S., Brown, T.L., 2001. Speech-based interaction with in- transportation systems. Human Factors and Ergonomics in Manufacturing 10
vehicle computers: the effect of speech-based e-mail on drivers’ attention to the (4), 395–408.
roadway. Human Factors 43 (4), 631–640. Yee, S., Nguyen, L., Green, P.A., Oberholtzer, J., Miller, B., 2007. Visual, auditory,
Lin, T.-W., Hwang, S.-L., Su, J.-M., Chen, W.-H., 2008. The effects of in-vehicle tasks cognitive, and psychomotor demands of real in-vehicle tasks (Technical report
and time-gap selection while reclaiming control from adaptive cruise control UMTRI-2006-20). Ann Arbor, MI: University of Michigan Transportation
(ACC) with bus simulator. Accident Analysis and Prevention 40 (3), 1164–1170. Research Institute.
Ohno, H., 2001. Analysis and modeling of human driving behaviors using adaptive Zheng, P., McDonald, M., 2005. Manual vs. adaptive cruise control – Can driver’s
cruise control. Applied Soft Computing 1 (3), 237–243. expectation be matched? Transportation Research Part C 13 (5-6), 421–431.