You are on page 1of 8

Effects of Adaptive Cruise Control

and Alert Modality on Driver Performance


John D. Lee, Daniel V. McGehee, Timothy L. Brown, and Dawn Marshall

Adaptive cruise control (ACC) is a rapidly emerging in-vehicle tech- vehicle-based sensors to estimate the distance and velocity relative to
nology that can enhance or degrade driving safety. A critical factor the vehicle ahead and then modulates the accelerator and service
governing the safety benefit of ACC concerns the driver’s ability to brakes to maintain a set time headway from the lead vehicle (1).
assume control of the vehicle in situations that exceed ACC capabili- Because ACC is designed as a convenience system rather than as a
ties. The effectiveness of various warning modalities for reengaging safety system, it does not engage the full braking potential of the
distracted drivers during severe braking situations that exceed ACC vehicle. In situations that require severe braking, the system alerts
capability was examined. Warnings that paired a visual icon with sound, the driver to the need to intervene to avoid striking the moving or
seat vibration, or brake pulsation or that combined all three cues were stationary vehicle (2, 3). This warning serves as a forward collision
compared. A total of 60 participants drove for 35 min in the National warning (FCW), and in combination with ACC, it could substan-
Advanced Driving Simulator. Drivers experienced two severe, four tially improve drivers’ ability to maintain safe speeds and headways
moderate, and eight mild braking situations. ACC could accommodate and to avoid collisions.
all but the two severe situations without driver intervention. It also Although offered as a convenience system, ACC may deliver an
provided a substantial benefit during mild events of lead vehicle brak- added safety benefit because it can provide a rapid response if the
ing, enabling drivers to maintain a longer, more consistent minimum lead vehicle slows. The ACC response time to a braking vehicle—
time to collision. Unlike performance in previous studies, ACC did not approximately 300 ms—can be substantially less than that of a
decrease safety during severe braking situations. The combination of typical driver. ACC also may provide indirect benefits in that ACC
a visual icon with sound, seat vibration, or brake pulsation led to braking is less abrupt and variable than that of a typical driver. ACC
slower brake reaction time in severe braking situations, but the group- can attenuate traffic disturbances that a driver might otherwise accen-
ing of all four warning strategies led to a similar minimum time to col- tuate (4–6). Such disturbances can contribute to crashes and induce
lision and maximum braking. In contrast to several previous studies, congestion. In one traffic simulation model, congestion occurred
these results suggest that drivers can effectively assume control when when 10% of the vehicles engaged ACC, but not when 20% of
warned that ACC braking authority has been exceeded. Additional vehicles engaged ACC (7). Because people do not always rely on
research is needed to identify the boundary conditions that specify automation appropriately, even relatively simple automation such as
when drivers can successfully intervene and retake control and ACC can undermine driving safety. In other domains, automation
whether a multimodal combination of cues can be crafted to speed with a high level of authority and little feedback has degraded event
rather than slow driver response. detection and slowed response (8–11). Similarly, poor feedback and
imprecise mental models of automation can lead to miscalibrated
trust in, increased complacency about, and inappropriate reliance on
New technology is changing the task of driving in ways that can system capability (12–14). As a form of automation that offers rela-
enhance or degrade driving safety. Technology that actively inter- tively little feedback and operates with a high level of authority, ACC
venes and controls the speed or direction of a vehicle may have the may cause similar problems and so may undermine driving safety. In
greatest effect on safety. Adaptive cruise control (ACC), augmented one simulator study, approximately one-third of drivers were unsuc-
braking, and steering assist systems all represent technology that cessful in assuming control after ACC had failed (15). Distractions
could actively intervene to augment human perceptual–motor capa- associated with cell phones and other in-vehicle technology may
bilities and enhance driving safety. Of these, ACC is becoming exacerbate this effect by encouraging drivers to rely on vehicle
widely available in production passenger vehicles. automation and neglect the driving task.
ACC operates much like conventional cruise control when Drivers may need alerts to signal the point at which they need to
no other vehicles are present in front of a driver. When an ACC- intervene. The type of alert could have an important influence on
equipped vehicle approaches slower-moving vehicles, ACC uses driver performance and acceptance. Potential alerts include auditory
cues, such as a tone or verbal notice; haptic cues, such as a seat
vibration, steering wheel jerk, or brake jerk; and visual displays,
J. D. Lee, Cognitive Systems Laboratory, Department of Mechanical and Industrial
such as a heads-up or icon located in the instrument cluster (16–18).
Engineering, 2130 Seamans Center; D. V. McGehee, Human Factors and Vehicle
Safety Research Program, Public Policy Center, 227 South Quad; and T. L. Brown Warnings often combine these cues. Little research has addressed
and D. Marshall, Human Factors Laboratory, National Advanced Driving Simulator warning modalities. Standards such as SAE J2399 simply describe
and Simulation Center, University of Iowa, 2401 Oakdale Boulevard, Iowa City, algorithmic issues and indicate that drivers should be warned when
IA 52242. the ACC’s braking authority is exceeded (3).
Transportation Research Record: Journal of the Transportation Research Board,
Of the possible warning cues, haptic cues are the most promis-
No. 1980, Transportation Research Board of the National Academies, Washington, ing and least investigated. Haptic cues may be more congruent
D.C., 2006, pp. 49–56. with expected driver response—a brake pulsation naturally draws

49
50 Transportation Research Record 1980

attention to the brake and has a greater degree of stimulus–response Experimental Design
compatibility than an auditory alert that has no natural association
with braking (19). This consideration may become increasingly A mixed between–within experimental design assessed the effect
important as alerts from in-vehicle information systems proliferate. of warning modalities and braking events. The warning mode was a
Another critical issue concerns how the alert modality affects between-subjects variable with four conditions: auditory alert, seat
driver response to inappropriate alerts, that is, those that are not vibration, brake pulsation, and a combination of all three. All warnings
related to a safety-relevant event. Haptic alerts might be particularly also included a visual alert.
likely to induce braking responses to inappropriate alerts, and pre- In a fifth condition, one group of participants drove without ACC
liminary evidence suggests some people may be particularly prone but had a FCW with an auditory and visual warning that was the
to such inappropriate responses; in one study, one participant in same as the auditory and visual warning of drivers with ACC. In this
a group of seven accounted for 40% of the inappropriate braking condition, braking events were matched to ACC conditions by grad-
responses (20). Currently produced vehicles use auditory and visual ually coupling the lead vehicle to the participant’s vehicle 5 s before
cues when situations exceed the ACC braking authority, and rela- the onset of severe and moderate events. The lead vehicle and the par-
tively little research has considered the potential benefit of haptic ticipant’s vehicle were coupled by programming the lead vehicle to
cues for collision-warning devices (21). assume and then maintain a 2-s headway just before the braking event.
The objective of this study is to assess drivers’ ability to tran- Providing no warning or other feedback when the braking authority
sition from ACC to manual control when warned with alerts of of the ACC has been exceeded is a logical alternate condition that
different modalities. This study compares headway maintenance might have been investigated in this experiment; however, this exper-
performance for drivers with and without ACC during mild, mod- imental condition was not considered because it is inconsistent with
erate, and severe braking events. Specifically, this research exam- existing standards (3).
ines the effectiveness of warning modalities (i.e., visual alert, The experiment also included event type, event severity, and
auditory alert, seat vibration, brake pulsation, and a combination response appropriateness as within-subject variables. Events in which
of all these) in supporting driver response to situations requiring the lead vehicle braked (braking lead vehicle) were differentiated
severe braking in two crash scenarios [lead vehicle brakes (referred from events in which the lead vehicle changed lanes to reveal a slow-
to as a braking lead vehicle) and an abrupt lane change by a lead moving vehicle (revealed lead vehicle). Response appropriate-
vehicle that reveals a slow-moving vehicle (referred to as a revealed ness refers to whether the ACC or FCW system responded to the
lead vehicle)]. A secondary objective is to assess the degree to traffic situation appropriately. For drivers with only FCW (no ACC),
which alert modality affects drivers’ compliance with inappropriate inappropriate responses were FCW warnings that occurred without a
alerts. corresponding need to brake. For drivers with ACC, inappropriate
responses included unnecessary mild and moderate deceleration by
the ACC. Similar to the drivers in the no-ACC condition, inappropri-
METHODS ate responses to severe events in the ACC condition included in-
appropriate braking and an FCW warning without a corresponding
Participants need to brake. These inappropriate warnings occurred once for each
driver. The inappropriate alerts occurred in response to overpasses,
Sixty people aged 30 to 50 years (M age = 41.7 years, SD = 42.0) and curves, and vehicles in the adjacent lane. Table 1 summarizes the
balanced by gender participated in this study. Participants were events that define the within-subject conditions. Two orders of events
screened so that they all drove at least 4,500 mi per year, held a valid were developed and counterbalanced such that one-half the subjects in
driver’s license, and used cruise control at least twice per month. each of the between-subjects conditions experienced one of the orders.
Participants completed a demographic questionnaire and received Moderate and severe events triggered a warning if drivers kept the
predrive instructions that took approximately 25 min to administer. ACC engaged. The warnings occurred 3.5 and 1.7 s after the start of
Next, participants drove a 6-min practice drive, followed by a 35-min the moderate and severe revealed lead vehicle events, and 2.0 s and
experimental drive. The total participation time for each person was 0.7 s after the start of the moderate and severe braking lead vehicle
approximately 105 min. Participants were paid $20/h with a total events. The start of the revealed lead vehicle event was defined by
compensation of approximately $40. the moment the lead vehicle began to change lanes to reveal the slow-

TABLE 1 Summary of Events Presented to Each Driver

Initial Time Initial Initial Lead Lead Duration of Lead


Event Headway Distance Vehicle Speed Vehicle Decel. Constant Vehicle
Event Severity (s) (ft) (mph) Decel. (g) Duration (s) Speed (s) Accel. (g) Times

Revealed lead vehicle Mild 3.2 305 45 n/a n/a 3.75 0.3 4
Moderate 3.2 305 30 n/a n/a 3.75 0.3 2
Severe 3.2 305 20 n/a n/a 6.0 0.3 1
Braking lead vehicle Mild 2.0 190 65 0.2 1.5 2.0 0.3 4
Moderate 2.0 190 65 0.4 2.5 2.5 0.3 2
Severe 2.0 190 65 0.7 2.25 3.0 0.3 1
Inappropriate response Mild 2.0 190 65 0.0 n/a n/a n/a 4
Moderate 2.0 190 65 0.0 n/a n/a n/a 2
Severe 2.0 190 65 0.0 n/a n/a n/a 1
Lee, McGehee, Brown, and Marshall 51

moving vehicle. The onset of the warnings highlights an important manual and ACC braking. The motion base generated longitudinal
difference in driver response and the ACC or FCW system. Drivers acceleration cues that matched those specified by the simulation of the
are relatively insensitive to the looming cues during the initial brak- vehicle dynamics. Specifically, if the driver pressed the brake and
ing of the lead vehicle (22, 23), whereas the cues associated with a decelerated at 0.4 g, the motion base produced acceleration cues of
lead vehicle lane change are relatively salient. In contrast, the FCW 0.4 g, that is, a 1:1 scaling of acceleration. During extreme braking,
responds quickly to the lead vehicle braking. An attentive driver nonlinear scaling ensured that commanded decelerations remained
might respond more quickly than the ACC or FCW during revealed within the limits of the motion base, which allowed approximately
lead vehicle but not braking lead vehicle events. 0.6 g of sustained deceleration. Lateral and vertical scaling were set at
In addition to the 14 events that required a response from the ACC 0.5 to accommodate the effects of the curves and leave the motion base
or the driver, drivers were exposed to seven instances in which the maximum flexibility in generating longitudinal acceleration cues.
ACC responded as if there were a traffic event, but there was none. Both practice and experimental drives were conducted on a rural
These inappropriate ACC responses involved mild or moderate freeway with two lanes of traffic in each direction and with a posted
braking and triggered the FCW alert in one instance. Drivers with speed limit of 65 mph. Traffic density was 18 to 35 vehicles/lane/mi.
only the FCW received a single inappropriate warning, correspond- The headway between vehicles was 1.2 to 2.6 s. This headway cor-
ing to the last line in Table 1. Overall, drivers were exposed to only responds to level of service B, where the traffic was relatively free-
two severe braking events and one inappropriate warning during the flowing but somewhat influenced by other vehicles (29). Drivers
35-min drive. were instructed to stay in the right lane and follow the vehicle ahead.
The auditory alert and visual display were the same for the no-ACC A car in the left lane shadowed the driver to discourage lane changes.
condition and the ACC-with-FCW condition. The visual display was Distraction tasks were triggered at preset locations along the
a high, heads-down display consisting of an icon of a collision between route. First, an audio message instructed the participant to press a
two cars from the driver’s perspective; it was triggered in the same button on the lower right of the center stack (console) and watch a
manner as in previous studies (24). The auditory alert was a tone sim- collocated digital display. The button press caused the display of a
ilar to the warning used in previous studies (25, 26). The volume of this random sequence of digits at a rate of 4 Hz. The driver’s task was to
was 80 A-weighted decibels (dBA), approximately 10 dB above the count the number of times the numeral 4 appeared. This visually
ambient noise of the vehicle; one cycle of the tone lasted 2.1 s. demanding distraction occurred 62 times during the drive. The mean
The haptic seat was a standard automotive seat modified to include time between occurrences was 30.4 s, with a standard deviation of
24 vibrating actuators, similar to those used in previous studies 15.4 s. Participants were told to drive normally and to attend to the
(17, 27). The actuators were configured to deliver vibrations that nondriving task as they might in a real driving situation.
started at the front of the seat the progressed to the rear. The pro- A button press associated with a distraction task also initiated the
gression took 2.25 s to complete and included a 500-ms pause with- events described in Table 1. Drivers experienced 14 events that
out vibration at the end of the cycle. The brake pulsation was a half required an ACC or manual braking response; these events occurred
sine wave with a peak magnitude of 0.015 g and lasted 400 ms. Each approximately 22.6% (14 out of 62) of the times the button was
warning was repeated if the conditions that had triggered the warning pressed. The frequency with which moderate and severe events
were still valid when the warning signal had completed its cycle. coincided with the button press was only 9.7% (6 out of 62). Link-
The ACC system used for this study [described by VanderWerf ing the event onset with the button press ensured that the driver was
et al. (28)] provides for both free drive and interaction with other at least moderately distracted at the onset of a braking event.
vehicles on the roadway. In free-drive mode, acceleration inputs
were directly proportional to the difference between the current
speed and the desired speed. When impeded by a slow-moving lead Procedure
vehicle, the ACC used a linear model that adjusted acceleration accord-
ing to the difference between actual and desired distance. Desired dis- Participants listened to a description of the experiment and con-
tance was determined on the basis of the desired time gap and the sented to participate. They then completed a demographic ques-
current speed. The algorithm is constrained to brake at no more than tionnaire that assessed driving habits. The experimenter described
0.25 g. In moderate and severe events, ACC braked at approximately the study details and provided a paper copy and recorded descrip-
0.25 g. This braking began approximately 300 ms after the event onset, tion of the in-vehicle technology that drivers would experience. The
simulating sensor and algorithm latencies that are present in actual number-counting activity was also described. The participants could
ACC systems. The onset of ACC braking provided drivers with an ask questions at any time.
additional cue regarding the possible need to intervene. The participants were escorted into the simulator, where they
received specific instructions about the simulator cab. Each began
with a short practice drive of ∼6 min. During the practice drive, par-
Simulator, Scenario, and Virtual Environment ticipants familiarized themselves with the vehicle and its handling
characteristics and were reminded how to use the ACC and FCW
The National Advanced Driving Simulator (NADS) was used for this systems. The practice drive included examples of the scenario
experiment. This simulator includes 360 × 40° view and a motion of braking lead vehicle and warning system activation. After the
base capable of replicating sustained accelerations of 0.6 g and vibra- practice drive, all participants completed a National Aeronautics
tions of 3 to 40 Hz. A sound system provides three-dimensional audi- and Space Administration (NASA) task load index (TLX) workload
tory cues that include wind and road noise as well as the sound of scale and trust scale.
other vehicles. The experiment was conducted in a 1996 Chevrolet During the main drive, participants drove along a rural highway
Malibu cab. for approximately 35 min. After completing the main drive, each
The NADS Dyna dynamics model paired with the motion base pro- participant completed a workload and trust survey (30–32) and a
vided participants with realistic acceleration cues associated with simulator sickness questionnaire (33). Each participant then was
52 Transportation Research Record 1980

escorted from the simulator and debriefed, completed a simulator benefit was primarily associated with braking lead vehicle events.
realism questionnaire and payment forms, and was escorted from Importantly, the scale differs substantially in these two figures. The
the building. minimum TTC is generally much shorter in revealed lead vehicle
events (Figure 1) than in braking lead vehicle events (Figure 2).
For the moderate braking lead vehicle event, post hoc tests show
RESULTS that no ACC condition has a significantly shorter minimum TTC
and that the brake pulsation cue leads to a significantly greater
All analyses were performed using the SAS general linear model minimum TTC (5.88 s) than the auditory cue (5.57 s). No differ-
(GLM) with a statistical significance criterion of P < 0.05. For all post ences for warning modality were significant in the moderate revealed
hoc evaluations, the Waller–Duncan k-ratio t-test was used to maxi- lead vehicle events, although a trend favored brake pulsation, sim-
mize the ability to find differences present in the data while providing ilar to the braking lead vehicle event. Among only severe events,
more control for Type I error than the standard t-test. Only four neither the modality of the alert nor the presence of ACC had a sta-
crashes occurred, and crash data were excluded from the analysis tistically significant effect on the minimum TTC [F(4,40) = 0.26;
because crashes were not concentrated in any particular condition. P > 0.05].
The first analysis excluded inappropriate ACC and FCW responses Although the minimum TTC during severe events was similar for
and included only those events listed in the first six rows of Table 1. all ACC alert modalities, brake reaction time was substantially slower
Because the numbers of occurrences were different during the drive for those drivers with combined auditory alert, seat vibration, and
for mild, moderate, and severe events, means were calculated for brake pulsation warning, for both revealed lead vehicle and braking
each event severity and each event type. The value analyzed for the lead vehicle events [F(4,50) = 2.91; P < 0.05]. Brake reaction time is
mild events represented the mean of four mild events, the value for defined as the time from the warning onset to initial depression of the
the moderate events represented the mean of two events, and the brake pedal. Post hoc test results indicated that the combination of
value for the severe events was the actual measure for the severe cues resulted in a slower response than the no-ACC and auditory-alert
event. The GLM model for this analysis included ACC conditions, conditions. The auditory and visual alerts (leftmost columns in Fig-
event severity, event type, and gender. ure 3) resulted in the fastest response, which did not depend on
The ACC alert modalities, along with event severity, event type, and whether drivers had ACC. Drivers responded to the warnings for
their interactions influenced the minimum time to collision (TTC). Not revealed lead vehicle events faster (0.58 s) than to those for brak-
surprisingly, more severe events were associated with smaller mini- ing lead vehicle events (1.23 s) or inappropriate warnings (1.22 s)
mum TTC [F(2,100) = 5,467.04; P < 0.001]. The revealed lead vehi- [F(2,49) = 33.4; P < 0.0001].
cle events also led to much smaller minimum TTC than the braking The alert modality had no significant effect on driver reaction to
lead vehicle events [F(2,100) = 5,734.22; P < 0.001]. inappropriate ACC and FCW responses. A detailed analysis of video
More interesting is the effect of the presence of the ACC system. recordings of drivers’ foot movements indicated that neither the degree
Figures 1 and 2 show that drivers with ACC had larger minimum to which the foot hovered over the brake [F(4,100) = 0.60; P > 0.05]
TTC during mildly and moderately severe situations; however, this nor the frequency with which drivers depressed the brake [F(4,100) =

6
Minimum time to collision (s)

0
n

e
C

C
d

d
io

ls

io

ls

io

ls
un

ne

un

ne

un

ne
AC

AC

AC
pu

pu

pu
at

at

at
So

bi

So

bi

So

bi
br

br

br
o

o
e

e
om

om

om
Vi

Vi

Vi
N

N
ak

ak

ak
C

C
Br

Br

Br

Mild Moderate Severe

FIGURE 1 Minimum TTC as function of ACC alert modality and event severity: revealed lead vehicle scenario.
Lee, McGehee, Brown, and Marshall 53

30

Minimum time to collision (s) 25

20

15

10

0
n

e
C

C
d

d
io

ls

io

ls

io

ls
un

ne

un

ne

un

ne
AC

AC

AC
pu

pu

pu
at

at

at
So

bi

So

bi

So

bi
br

br

br
o

o
e

e
om

om

om
Vi

Vi

Vi
N

N
ak

ak

ak
C

C
Br

Br

Br
Mild Moderate Severe

FIGURE 2 Minimum TTC as function of ACC alert modality and event severity: scenario of braking lead vehicle.

0.56; P > 0.05] was related to alert modality. Most (78.3%) drivers time the button was pressed [F(4,55) = 1.1; P > 0.05] or for the 3 s fol-
moved their foot or hovered their foot over the brake in response to the lowing the button press [F(4,81) = 1.81; P > 0.05]. Overall, drivers
inappropriate collision warning, but only 56.6% actually depressed the kept their eyes on the road for approximately 68% of the 3 s after the
brake. Similarly, the maximum braking of drivers in all alert modality button press, regardless of whether they had ACC and independent of
conditions was similar [F(4,50) = 0.45; P > 0.05]. the alert modality [F(4,88) = 0.40; P > 0.05].
ACC presence did not affect engagement with the secondary task, The mild and moderate events did compel some drivers to
as measured by the amount of time drivers looked at the secondary respond by either moving their foot or applying the brake, even
task display from the onset of the instruction to begin the task to the though these responses were not necessary for drivers with ACC.

1.40
Brake Reaction Time to Warning (sec)

1.20

1.00

0.80

0.60

0.40

0.20

0.00
No ACC Auditory Seat Vibration Brake Pulse Combination

FIGURE 3 Brake reaction time to severe braking situations as a function of ACC alert modality.
54 Transportation Research Record 1980

On average, drivers responded to 10% of the mild events, 45% of Another explanation for the apparent benefit of ACC is that
the moderate events, and 95% of the severe events. Although ACC it augmented drivers’ responses by initiating a braking response
could accommodate the mild and moderate events, the rate of approximately 0.5 to 1.5 s before the warning onset and by contin-
response for drivers with ACC was approximately the same as for uing the response until the driver responded. In addition, the ACC
drivers without ACC. braking response and associated deceleration cues may be critical
in reengaging the driver in the control process. Studies in fixed-base
driving simulators do not provide these deceleration cues and may
DISCUSSION overestimate driver difficulty in reengaging and responding to crit-
ical braking events. Future research should assess the degree to
ACC helped drivers maintain a larger safety margin, as measured by which the profile of the onset of ACC braking influences driver
the minimum TTC, during mild and moderate braking lead vehicle response to critical situations.
events. Because ACC helped maintain a larger safety margin during In generalizing the results of this study, it is important to con-
less severe braking events, it may have important indirect benefits sider that drivers were exposed to a relatively large number of
(e.g., affecting other drivers and the overall traffic flow rather than events. However, only two events required a response in the ACC
the crash likelihood of the driver using it). condition and therefore substantially surprised drivers. Because
An important overall safety benefit of ACC may be its ability these events always were triggered when the driver initiated the
to dampen disruptions in traffic flow by maintaining a more con- distraction task, drivers may have associated the distraction task
sistent TTC in response to small disturbances. Small disturbances with traffic events. However, the majority of the distraction tasks
can transform free-flowing traffic into congestion and create were not associated with an event. As a result, drivers did not
severe braking situations that can endanger drivers (34, 35). This prepare to brake every time they pressed the button—90% of the
study complements previous traffic simulation studies and shows drivers did not respond to mild events, not even with a slight foot
how ACC might dampen disruptions associated with lead vehicle movement. If drivers had been cued by the secondary task to
braking and abrupt lane changes. The benefit of this effect of ACC expect a lead vehicle change, then this number would have been
on traffic flow stability is considerable (36); it can reduce the much higher. At the same time, drivers were faced with an active
number of hazardous lane changes and even decrease pollution driving environment that tended to keep them engaged in the
levels (4). driving task.
ACC resulted in a longer, more consistent minimum TCC, mainly The response to events that could have been accommodated
in mild and moderate braking lead vehicle events. One explanation by ACC was similar for drivers with and without ACC, suggest-
is that the visual information available to drivers when the lead
ing that drivers carefully monitored ACC performance. Fewer events
vehicle brakes—increasing the visual angle of the lead vehicle—
and more experience with ACC might have led to disengagement
is a much less salient cue than the abrupt threat onset that occurs
with the driving task, making it more difficult for drivers to reen-
when the lead vehicle changes lanes (22). Because ACC responds
ter the control loop. At the same time, fewer events may have led
immediately to dangers that drivers may not appreciate, ACC
to a greater degree of surprise for drivers with and without ACC,
might provide a particularly large safety benefit in braking lead
and in such a situation the benefit of ACC and FCW may have
vehicle situations.
been greater.
The apparent benefit of the ACC in this study contrasts with
Another challenge in generalizing these results is the degree
other studies that have shown declining performance with such
systems (37 ). One explanation for this difference is that drivers in to which specific ACC and FCW characteristics affect driver response.
this study received warnings for all braking events to which a ACC systems differ with respect to control algorithm and maximum
response was required. In other studies, ACC failed without warn- braking and could increase or decrease driver engagement and corre-
ing. This study focused on situations that exceeded the braking spondingly increase or decrease reaction times in response to road-
authority of the ACC, in contrast to studies that considered the way events. These differences might have important effects on how
effect of sensor failures. For example, in one study, more than 25% well ACC and FCW enhances driving safety. More research is needed
of drivers failed to intervene when a sensor failure caused the ACC to to explore how drivers respond to situations of greater surprise and
accelerate into a car ahead (37 ). greater disengagement from the driving task.
These results point to two distinct failure modes that merit atten- Substantial research has demonstrated the benefit of using redun-
tion in ACC evaluation. The first represents limits of ACC brak- dant sensory channels to convey information (38, 39). For example,
ing authority for which appropriate alerts have been developed. responses to a combined visual, auditory, and tactile stimulus were
This study shows that even distracted drivers can resume control faster than those to stimuli composed of two modalities, which were
effectively if they are alerted to the inability of the ACC to accom- faster than those to single-modality stimuli (40). Similarly, adding
modate an evolving traffic situation. The second represents un- visual cues to an auditory warning enhanced driver response (41).
expected failures for which alerts are not typically provided (37 ). This redundancy gain suggests that the combination of visual alert,
Drivers respond poorly to these situations, often failing to inter- auditory alert, seat vibration, and brake pulsation would support the
vene in a timely manner. Such failures occur rarely and may be fastest response in the current study; however, it was not the case.
underrepresented in field tests because field-test vehicles tend to The multimodal combination was more than 400 ms slower than the
be newer, better maintained, and less prone to sensor failures. auditory and visual alert.
Future research assessing the benefit of providing drivers with This result is not unique. In the context of human–computer inter-
real-time information about sensor performance may be critical to action, feedback that combined auditory, haptic, and visual informa-
avoid this second failure mode, which seems to be a particularly tion performed worse than the bimodal visual and haptic combination
potent threat to driving safety. (42). In the context of patient monitoring, a redundant combination of
Lee, McGehee, Brown, and Marshall 55

auditory and visual information performed worse than either an audi- suraman, eds.), Lawrence Erlbaum Associates, Hillsdale, N.J., 1994,
tory or a visual alert alone (43). Driver response to a lane-departure pp. 45–49.
13. Lee, J. D., and N. Moray. Trust, Self-Confidence, and Operators’ Adapta-
collision warning had a similar effect, in which the multimodal
tion to Automation. International Journal of Human–Computer Studies,
warning also resulted in slower responses (44). Although generally Vol. 40, No. 1994, pp. 153–184.
beneficial, redundancy gain is thus not universal. 14. Lee, J. D., and K. A. See. Trust in Technology: Designing for Appropri-
One possible explanation is that multimodal warnings can be per- ate Reliance. Human Factors, Vol. 46, No. 1, 2004, pp. 50–80.
ceived as a single cue or as a set of cues. Response to a multimodal 15. Stanton, N. A., M. Young, and B. McCaulder. Drive-by-Wire: The
Case of Driver Workload and Reclaiming Control with Adaptive
warning may be enhanced when the warning is perceived as a single
Cruise Control. Safety Science, Vol. 27, No. 2, 1997, pp. 149–159.
cue but degraded when it is perceived as multiple cues. Understand- 16. Hirst, S., and R. Graham. The Format and Presentation of Collision
ing how multimodal warnings can capitalize on cross-modal inatten- Warnings. In Ergonomics and Safety of Intelligent Driver Interfaces
tion may benefit from recent findings in neuropsychology (45, 46). (I. Noy, ed.), Lawrence Erlbaum Associates, Mahwah, N.J., 1997,
Matching the time and frequency profiles may help create a single- pp. 203–219.
17. Lee, J. D., J. D. Hoffman, and E. Hayes. Collision Warning Design to Mit-
cue gestalt that drivers can quickly process. Achieving effective com-
igate Driver Distraction. Proc., CHI2004: Conference on Human Factors
munication between ACC and driver may depend on identifying the in Computing Systems, Vienna, Austria, Association for Computing
characteristics of multimodal signals that must be matched to form Machinery, New York, 2004, pp. 65–72.
such a gestalt. 18. Graham, R. Use of Auditory Icons as Emergency Warnings: Evaluation
Within a Vehicle Collision Avoidance Application. Ergonomics, Vol. 42,
No. 9, 1999, pp. 1233–1248.
19. Wickens, C. D., J. D. Lee, Y. Liu, and S. E. Gordon. An Introduction to
ACKNOWLEDGMENT Human Factors Engineering, 2nd ed. Longman, New York, 2003.
20. Tijerina, L., S. Johnston, E. Parmer, H. A. Pham, M. D. Winterbottom, and
This research was sponsored by the National Highway Traffic F. S. Barickman. Preliminary Studies in Haptic Displays for Rear-End
Safety Administration. Collision Avoidance System and Adaptive Cruise Control Applications.
National Highway Transportation Safety Administration, Washington,
D.C., 2000.
21. Ho, C., H. Z. Tan, and C. Spence. Using Spatial Vibrotactile Cues to
REFERENCES Direct Visual Attention in Driving Scenes. Transportation Research Part
F: Traffic Psychology and Behaviour, Vol. 8, No. 2005, pp. 397–412.
1. Fancher, P., Z. Bareket, S. Bogard, C. MacAdam, and R. Ervin. Tests 22. Hoffmann, E. R., and R. G. Mortimer. Scaling of Relative Velocity
Characterizing Performance of an Adaptive Cruise Control System. In Between Vehicles. Accident Analysis and Prevention, Vol. 28, No. 4,
Object Detection, Collision Warning and Avoidance Systems (R. K. 1996, pp. 415–421.
Jurgen, ed.), Society of Automotive Engineers, Warrendale, Pa., 1998, 23. Hoffmann, E. R., and R. G. Mortimer. Drivers Estimates of Time to
pp. 273–282. Collision. Accident Analysis and Prevention, Vol. 26, No. 4, 1994,
2. McGehee, D. V., D. J. LeBlanc, R. J. Kiefer, and J. Salinger. Human Fac- pp. 511–520.
tors in Forward Collision Warning Systems: Operating Characteris- 24. Automotive Collision Avoidance System Field Operational Test: Final
tics and User Interface Requirements. Society of Automotive Engineers, Program Report. National Highway Traffic Safety Administration,
Warrendale, Pa., 2002. Washington, D.C., 2005.
3. Sayer, J. R. Adaptive Cruise Control (ACC) Operating Characteristics 25. Tan, A., and N. Lerner. Multiple Attribute Evaluation of Auditory Warn-
and User Interface: Standard J2399. Society of Automotive Engineers, ing Signals for In-Vehicle Crash Warning Systems. National Highway
Warrendale, Pa., 2003. Transportation Safety Administration, Washington, D.C., 1995.
4. Ioannou, P. A., and M. Stefanovic. Evaluation of ACC Vehicles in 26. Lee, J. D., D. V. McGehee, T. L. Brown, and M. L. Reyes. Collision
Mixed Traffic: Lane Change Effects and Sensitivity Analysis. IEEE Warning Timing, Driver Distraction, and Driver Response to Immi-
Transactions on Intelligent Transportation Systems, Vol. 6, No. 1, 2005, nent Rear-End Collisions in a High-Fidelity Driving Simulator.
pp. 79–89. Human Factors, Vol. 44, No. 2, 2002, pp. 314–334.
5. Bose, A., and P. A. Ioannou. Analysis of Traffic Flow with Mixed Man- 27. Lee, J. D., E. M. Hayes, E. E. Wiese, and D. V. McGehee. Rear-End
ual and Semiautomated Vehicles. IEEE Transactions on Intelligent Collision Avoidance Systems: Single and Graded Warnings with Haptic,
Transportation Systems, Vol. 4, No. 4, 2003, pp. 173–188. Auditory, and Visual Warnings. National Highway Traffic Safety
6. Vahidi, A., and A. Eskandarian. Research Advances in Intelligent Col- Administration, Washington, D.C., 2002.
lision Avoidance and Adaptive Cruise Control. IEEE Transactions on 28. VanderWerf, J., S. Shladover, N. Kourjanskaia, M. Miller, and H. Krish-
Intelligent Transportation Systems, Vol. 4, No. 3, 2003, pp. 143–153. nan. Modeling Effects of Driver Control Assistance Systems on Traffic. In
7. Davis, L. C. Effect of Adaptive Cruise Control on Traffic Flow. Physical Transportation Research Record: Journal of the Transportation Research
Review E, Vol. 69, No. 6, 2004, pp. 066110-1–066110-8. Board, No. 1748, TRB, National Research Council, Washington, D.C.,
8. Endsley, M. R., and E. O. Kiris. The Out-of-the-Loop Performance 2001, pp. 167–174.
Problem and Level of Control in Automation. Human Factors, Vol. 37, 29. Highway Capacity Manual. TRB, National Research Council, Wash-
No. 2, 1995, pp. 381–394. ington, D.C., 2000.
9. Wickens, C. D., and C. Kessel. Failure Detection in Dynamic Systems. 30. Rotter, J. B. A New Scale for the Measurement of Interpersonal Trust.
In Human Detection and Diagnosis of System Failures (J. Rasmussen Journal of Personality, Vol. 35, No. 4, 1967, pp. 651– 665.
and W. B. Rouse, eds.), Plenum Press, New York, 1981, pp. 155–169. 31. Bisantz, A. M., and Y. Seong. Assessment of Operator Trust in and
10. Sarter, N. B., and D. D. Woods. “Strong, Silent, and ‘Out-of-the-Loop’ ”: Utilization of Automated Decision-Aids Under Different Framing
Properties of Advanced (Cockpit) Automation and Their Impact on Conditions. International Journal of Industrial Ergonomics, Vol. 28,
Human-Automation Interaction. Technical Report CSEL 95-TR-01. No. 2, 2001, pp. 85–97.
Cognitive Systems Engineering Laboratory, Ohio State University, 32. Hart, S. G., and L. E. Staveland. Development of NASA-TLX (Task
Columbus, 1995. Load Index): Results of Experimental and Theoretical Research. In
11. Sarter, N. B., and D. D. Woods. Team Play with a Powerful and Inde- Human Mental Workload (P. A. Hancock and N. Meshkati, eds.), North
pendent Agent: Operational Experiences and Automation Surprises on Holland, Amsterdam, Netherlands, 1988, pp. 139–183.
the Airbus A-320. Human Factors, Vol. 39, No. 4, 1997, pp. 553–569. 33. Kennedy, R. S., N. E. Lane, K. S. Berbaum, and M. G. Lilienthal. Sim-
12. Parasuraman, R., M. Mouloua, and R. Molloy. Monitoring Automation ulator Sickness Questionnaire: An Enhanced Method for Quantifying
Failures in Human-Machine Systems. In Human Performance in Auto- Simulator Sickness. The International Journal of Aviation Psychology,
mated Systems: Current Research and Trends (M. Mouloua and R. Para- Vol. 3, No. 3, 1993, pp. 203–220.
56 Transportation Research Record 1980

34. Kerner, B. S. Theory of Breakdown Phenomenon at Highway Bottle- 41. Belz, S. M., G. S. Robinson, and J. G. Casali. A New Class of Auditory
necks. In Transportation Research Record: Journal of the Transporta- Warning Signals for Complex Systems: Auditory Icons. Human Factors,
tion Research Board, No. 1710, TRB, National Research Council, Vol. 41, No. 4, 1999, pp. 608–618.
Washington, D.C., 2000, pp. 136–144. 42. Vitense, H. S., and J. A. Jacko. Multimodal Feedback: An Assessment of
35. Kerner, B. S. Synchronized Flow as a New Traffic Phase and Related Performance and Mental Workload. Ergonomics, Vol. 46, No. 1-3, 2003,
Problems for Traffic Flow Modeling. Mathematical and Computer pp. 68–87.
Modelling, Vol. 35, No. 5-6, 2002, pp. 481–508. 43. Seagull, J. F., C. D. Wickens, and R. G. Loeb. When Is Less More?
36. Li, P. Y., and A. Shrivastava. Traffic Flow Stability Induced by Con- Attention and Workload in Auditory, Visual, and Redundant Patient-
Monitoring Conditions. Proc., Human Factors and Ergonomics Society
stant Time Headway Policy for Adaptive Cruise Control Vehicles.
45th Annual Meeting, Minneapolis, Minn., Human Factors and Ergo-
Transportation Research Part C: Emerging Technologies, Vol. 10,
nomics Society, Santa Monica, Calif., 2001, pp. 1395–1399.
No. 4, 2002, pp. 275–301. 44. Tijerina, L., J. L. Jackson, D. Pomerleau, R. A. Romano, and A. O.
37. Stanton, N. A., and M. S. Young. Vehicle Automation and Driving Petersen. Driving Simulator Tests of Lane Departure Collision Avoid-
Performance. Ergonomics, Vol. 41, No. 7, 1998, pp. 1014–1028. ance Systems. Proc., ITS America 1996 Annual Meeting, Houston, Tex.,
38. Todd, J. W. Reaction Time to Multiple Stimuli. Archives of Psychology, Intelligent Transportation Society of America, Washington, D.C., 1996,
Vol. 3, No. 1912, pp. 1–65. pp. 636– 648.
39. Nickerson, R. S. Intersensory Facilitation of Reaction Time: Energy Sum- 45. Driver, J., and C. Spence. Cross-Modal Links in Spatial Attention.
mation or Preparation Enhancement? Psychological Review, Vol. 80, Philosophical Transactions of the Royal Society of London Series B,
No. 1973, pp. 489–509. Biological Sciences, Vol. 353, No. 1373, 1998, pp. 1319–1331.
40. Diederich, A., and H. Colonius. Bimodal and Trimodal Multisen- 46. Spence, C. Multisensory Attention and Tactile Information-Processing.
sory Enhancement: Effects of Stimulus Onset and Intensity on Reac- Behavioural Brain Research, Vol. 135, No. 1-2, 2002, pp. 57–64.
tion Time. Perception and Psychophysics, Vol. 66, No. 8, 2004,
pp. 1388–1404. The Vehicle User Characteristics Committee sponsored publication of this paper.

You might also like