Dow
nloa
ded
By:
[dbk
aber
@eo
s.nc
su.e
du] A
t: 19
:30
11 J
une
2007
Situation awareness and driving performancein a simulated navigation task
R. MA{ and D. B. KABER*{
{AREVA NP Inc., Charlotte, North Carolina, USA
{Edward P. Fitts Department of Industrial & Systems Engineering,
North Carolina State University, Raleigh, North Carolina, USA
The objective of this study was to identify task and vehicle factors that may
affect driver situation awareness (SA) and its relationship to performance,
particularly in strategic (navigation) tasks. An experiment was conducted to
assess the effects of in-vehicle navigation aids and reliability on driver SA and
performance in a simulated navigation task. A total of 20 participants drove
a virtual car and navigated a large virtual suburb. They were required to
follow traffic signs and navigation directions from either a human aid via a
mobile phone or an automated aid presented on a laptop. The navigation aids
operated under three different levels of information reliability (100%, 80%
and 60%). A control condition was used in which each aid presented a
telemarketing survey and participants navigated using a map. Results
revealed perfect navigation information generally improved driver SA and
performance compared to unreliable navigation information and the control
condition (task-irrelevant information). In-vehicle automation appears to
mediate the relationship of driver SA to performance in terms of operational
and strategic (navigation) behaviours. The findings of this work support
consideration of driver SA in the design of future vehicle automation for
navigation tasks.
Keywords: In-vehicle automation; Automation reliability; Situation aware-
ness; Driving performance; Navigation aiding
1. Introduction
Driving is considered to be a complex task requiring perception, comprehension and
projection of states of the roadway environment, as well as decision making on courses
of action and execution of driving behaviours. Situation awareness (SA) encompasses
three of these aspects of driving performance and has been found to be critical to
decision making in complex tasks (Endsley 1995a). Endsley (1995a) defined SA as: ‘the
*Corresponding author. Email: [email protected]
Ergonomics
Vol. 50, No. 8, August 2007, 1351–1364
ErgonomicsISSN 0014-0139 print/ISSN 1366-5847 online ª 2007 Taylor & Francis
http://www.tandf.co.uk/journalsDOI: 10.1080/00140130701318913
Dow
nloa
ded
By:
[dbk
aber
@eo
s.nc
su.e
du] A
t: 19
:30
11 J
une
2007
perception of the elements in the environment within a volume of time and space, the
comprehension of their meaning, and the projection of their status in the near future’.
Ma and Kaber (2005) summarized underlying factors in driver SA, including navigation
knowledge, environment and interaction knowledge, spatial orientation knowledge and
vehicle status knowledge. They suggested an integration of these forms of knowledge in
a model of driver information processing to achieve accurate SA. However, there have
been few empirical studies of how drivers may form these types of knowledge for SA
and what task, environment or system features, such as in-vehicle automation, play a
role.
Ward (2000) and Matthews et al. (2001) related levels of SA defined by Endsley,
including perception (Level 1 SA), comprehension (Level 2 SA) and projection (Level 3
SA), to specific driving tasks or behaviours, including operational, tactical and strategic
behaviour. For example, in operational driving tasks, drivers are engaged in actions upon
vehicle actuators in order to maintain stable control. Ward (2000) and Matthews et al.
(2001) said operational driving tasks, including steering and braking responses, primarily
require Level 1 SA. Matthews et al. (2001) said: ‘Level 2 SA may (also) be involved, if
driving processes generate error messages’. For example, a driver may comprehend the
rate of vehicle deceleration to be insufficient (Level 2 SA) for a particular perceived
stopping distance (Level 1 SA). In tactical driving tasks, Matthews et al. observed that
there is a high requirement for Levels 1 and 2 SA to, for example, facilitate safe
manoeuvring of a vehicle in traffic by judging and comparing lane positions. Level 3 SA is
also relevant to such interactive driving situations where there is a high requirement for
near-term projection of changes in the driving course and traffic patterns. In strategic
driving tasks, when navigation plans are being formulated, there is also a high
requirement for Level 3 SA.
At this point in research on driver SA, there is a need to identify operator, task,
environment and system factors that may mediate the role of the various levels of SA in
specific driving behaviours, including operational, tactical and strategic behaviours, in
following tasks, navigation tasks, etc. As an example, Ma and Kaber (2005) investigated
the effects of adaptive cruise control (ACC–see also Stanton and Young 2005) and mobile
phone use on driver SA and performance in a lead-car following task primarily involving
operational behaviours. Their results revealed that SA improved with the use of ACC and
decreased with use of the mobile phone under normal driving conditions. They also found
significant positive correlations between SA and aspects of driving performance. They
observed that all levels of SA (perception, comprehension and projection) appear to have
an impact on operational driving behaviours in following tasks and, consequently,
performance. In general, the use of ACC and mobile phones may be mediating factors in
linkages of SA to specific driving behaviours and performance.
Additional empirical work is needed to identify other in-vehicle system factors that
may be influential in driver SA and to describe the role of each level of SA in performance
of tactical and strategic driving behaviours or more complicated driving tasks (Walker
et al. 2006). Links between Levels 1 and 2 SA and tactical behaviour are supported by
Matthews et al.’s work, as well as a link between Level 3 SA and strategic behaviour. For
example, the use of in-vehicle automated navigation aids has become prevalent for
driving in new environments, or to unknown destinations, to support driver decision
making. However, like the use of mobile phones while driving, navigation aid use
represents a secondary task that has the potential to decrease driver attention to the
roadway and SA, particularly Level 1 (perception). This may in turn lead to performance
problems. That is, the aid may facilitate achieving high-level SA for navigation but may
1352 R. Ma and D. B. Kaber
Dow
nloa
ded
By:
[dbk
aber
@eo
s.nc
su.e
du] A
t: 19
:30
11 J
une
2007
also undermine attention to operational or tactical driving behaviours (e.g. braking,
passing, etc.) at certain points in the driving task.
In order to prevent automation-induced performance problems (e.g. Parasuraman and
Riley 1997) in lower-level (operational and tactical) driving behaviours, the reliability and
manner of delivery of navigation information is also critical in the design of in-vehicle
systems. There are few studies that have systematically varied automation reliability
levels in investigating human performance in complex systems control and results are
mixed. Dzindolet et al. (2001) required participants to view slides of battlefield terrain
and to indicate the presence or absence of a camouflaged soldier through the assistance of
an automated decision aid. Their results suggested that operators were insensitive to
differences in automation reliability. Wiegmann et al. (2001) examined the effect of
different levels of, and changes in, automation reliability on user performance with
automated diagnostic aids and found sensitivity to the different levels of aid reliability.
The objectives of the current research were to investigate the potential mediating effect
of in-vehicle navigation aids and aid reliability on the relationship of driver SA (at
multiple levels) with performance in a strategic driving and navigation task.
2. Experiment
2.1. Task
This study used a ‘home-grown’ virtual reality-based driving simulator to present a suburb
navigation and driving task in which drivers were only required to deal with following
traffic (i.e. there was no traffic in front of the user car). The simulation ran on a high-
performance graphics workstation and was presented to users through a stereoscopic
display. Participants viewed the 3-D display of the virtual driving environment and
manually controlled a virtual car (sport truck) to perform the navigation task. User inputs
to the simulation occurred through realistic automobile control interfaces, including a
physical steering wheel and acceleration and brake pedals (see figure 1). The simulation
was limited in fidelity in that it did not provide kinesthetic cues to drivers, as is possible in
full-motion simulators. This is important because it may have reduced driver sense of
vehicle feedback, which has been found to be relevant to speed and turning control
(Walker et al. 2006).
Participants were required to drive from a virtual freeway exit through the suburb and
to arrive at a pre-identified destination. The destination was located four (suburban)
blocks away from the freeway exit. There were three construction sites on the suburb
roadways and participants were required to make five or more turns to navigate the
sites and reach the destination. Figure 2 presents a map of the suburb area with street
names and the destination labelled. There were different types of traffic signs along
the streets, including ‘speed limit’ (30 or 45 mph), ‘stop’ signs and street name signs.
Participants were to obey all traffic signs and follow navigation aiding information
provided by either a human or automation aid without deviating from the directions in
their driving.
In another study, Dzindolet et al. (2002) found that humans exhibited a significant bias
towards the use of automated aids in terms of complex systems control over other human
aids. The automated aids were generally perceived as being more reliable for performance
than a human aid. With this in mind, we also wanted to see if the source of the navigation
aiding information mediated driver achievement of SA and strategic task/navigation
performance. Half of the participants (n¼ 10) received navigation instructions from a
Situation awareness and driving performance 1353
Dow
nloa
ded
By:
[dbk
aber
@eo
s.nc
su.e
du] A
t: 19
:30
11 J
une
2007
Figure 1. Experiment equipment setup.
Figure 2. Paper map of the suburban area with street names and destination.
1354 R. Ma and D. B. Kaber
Dow
nloa
ded
By:
[dbk
aber
@eo
s.nc
su.e
du] A
t: 19
:30
11 J
une
2007
remote experimenter via a Motorola T720 mobile phone (Motorola Inc.). The other half
(n¼ 10) received navigation information from an automated aid presented on an IBM
laptop placed directly adjacent to the primary virtual reality simulation display (see figure
1). In all navigation trials, participants had a paper map, which showed the entire suburb
navigation area. The participants could use the map to assess the reliability of the
navigation aid information or the accuracy of the route information during trials.
2.2. Variables
The independent variables in the experiment included the navigation aid type (i.e. human
or automation), which was manipulated between-subjects, and the level of navigation aid
reliability manipulated as a within-subjects variable, including 100%, 80% and 60%
reliable conditions, as well as a control condition involving a telemarketing survey
delivered through the navigation aids. The navigation aiding provided drivers with
turning information (street names) and speed limit information. All of the information
was correct according to the street signage; however, the overall accuracy of the route
information varied across aid reliability conditions in terms of the number of turns and
the total elapsed driving time. For example, if the minimum number of turns in an
optimal route from the freeway exit to the destination was five, and an aid provided
correct directions for only three of these turns, at least two corrective turns would need to
be added to the optimal route (for a total of seven turns) to reach the destination.
Consequently, the aid would only be 60% reliable in presenting correct turns on the
optimal route. In the experiment, the 100% reliable navigation condition required drivers
to make five turns and they spent an average of 11.7 min in reaching the destination from
the freeway exit. The 80% reliable navigation condition required drivers to make six
turns and average navigation time was 14.2 min. The 60% reliable navigation condition
required seven turns and an average driving time of 15.1 min.
Under the telemarketing survey condition, the paper copy of the map was marked with
an optimal route for navigation. The telemarketing survey was communicated by the
human through the mobile phone or by the automation aid through the laptop display.
Participants were required to answer the survey verbally (questions about shopping in the
Raleigh, North Carolina area) while an experimenter recorded their responses in writing.
The participants assigned to human or automation aiding followed the same trial orders
in terms of exposure to the reliability conditions. They experienced the levels of aid
reliability in descending order, followed by the telemarketing survey condition. This was
necessary to determine if any systematic change occurred in driver SA and performance
as the level of aid reliability degraded. Based on Dzindolet et al.’s (2002) finding, the
authors also wanted to determine if changes in SA with decreasing reliability might be
more pronounced for human aids vs. automated aids.
The dependent variables included driving performance, assessed in terms of the
number of errors drivers made in following the directional advice of a navigation aid and
their consistency in speed control in the suburb area (relative to posted limits). During the
navigation task, drivers were instructed to adhere to the directions of the aids regardless
of whether they knew the aid was unreliable; that is, the route information was
inaccurate. If, for example, the automated aid advised a driver to turn left at an
intersection and the driver turned right, an error count of one was recorded. The mean
number of errors was calculated across all trials for each navigation aid by reliability
condition. This approach was necessary, in part, to ensure that the performance measures
were sensitive. If participants were permitted to drive freely in the virtual suburb, it would
Situation awareness and driving performance 1355
Dow
nloa
ded
By:
[dbk
aber
@eo
s.nc
su.e
du] A
t: 19
:30
11 J
une
2007
not have been possible to assess the accuracy with which they perceived and executed
instructions from the aids, nor would it have been possible to completely assess the effect
of aid reliability on driver SA and performance.
Driver SA was assessed at the end of each trial using an adaptation of the SA global
assessment technique (SAGAT; Endsley 1995b). Nine queries were selected from a larger
database of queries and three questions targeting each level of driver SA (perception,
comprehension and projection) were presented. Participants were asked to recall, for
example, signs they passed during driving, when they passed certain signs or streets, etc.
They were also asked to project the drive time to the destination and possible optimal
routes to reach the destination. (Appendix 1 presents an example set of SA queries posed
to test drivers.) All queries were presented on paper and participants responded in
writing. Their responses were graded on the basis of the ‘ground truth’ of the simulation
recorded by the virtual reality system during each trial. Driver percent correct responses
to all queries were calculated as a measure of overall SA. The percentage of correct
responses to Levels 1, 2 and 3 SA queries were also determined in order to associate
changes in any of the specific levels with navigation performance changes.
One limitation of posing SA queries at the close of trials is that participants may need
to recall specific information on the driving environment from their earlier performance.
Therefore, the measure may be biased by subject recall ability. However, if SAGAT
freezes were used during the navigation trials, as in the original SAGAT methodology
developed by Endsley (1995b), there may have been disruptions to driver navigation
performance, influencing the other response measures. These types of trade-offs are
typical in the use of direct, objective measures or SA, such as SAGAT, and are often
evaluated based on the dynamics of the task under study (e.g. Kaber et al. 2005).
2.3. Participants and procedures
A total of 20 students were recruited for the study from the North Carolina State
University student population and they were required to have at least 3 years of driving
experience to participate. The average age of participants was 28.1 years and the sample
had an average of 8.5 years of driving experience.
Before participants began the test trials, they were provided with training on how to
control the virtual car in the simulator and how to maintain the vehicle on the virtual
roadway using the physical steering wheel, accelerator and brake pedals. In general, the
responsiveness of the simulator was very much like that of a real vehicle. The training
simulation environment presented no traffic signs or street name signs. Subsequently,
participants were provided with a practice session in which the simulation environment
remained the same. Drivers were shown the paper map of the suburb in advance and they
were informed of the location of the driving destination. They had access to the paper map
throughout the practice session. At the end of the suburb navigation, participants were
required to answer example SA questions. The practice session was intended to account
for potential roadway learning effects across the navigation aid reliability conditions (the
within-subjects variable settings). However, it is possible that some additional learning of
the simulated roadways may have occurred during the actual experiment.
Participants completed four test trials. They were instructed that the task was to
involve navigation under normal driving circumstances and had no reason to expect
hazardous conditions or otherwise. During the navigation task, the aids presented
information to drivers as they approached each street intersection or decision point.
A ‘wizard-of-oz’ technique was used, in which an experimenter, observing participant
1356 R. Ma and D. B. Kaber
Dow
nloa
ded
By:
[dbk
aber
@eo
s.nc
su.e
du] A
t: 19
:30
11 J
une
2007
driving performance on a separate remote monitor, called the participant on the mobile
phone, or the experimenter controlled the automated navigation aid display to deliver
specific driving direction information at the ‘right’ time. Figure 3 presents an example
script read by the human aid and an example automated aid display. There was a total of
six to eight mobile phone calls or six to eight automated aid display changes (depending
upon the navigation aid reliability) during each test trial. Participants were informed in
advance of the trials that the aids might not be completely reliable in the information
presented, but they were not informed of the specific reliability levels.
2.4. Detailed hypotheses
Higher reliability navigation aiding was expected to facilitate better driving performance.
It was also expected that presentation of the telemarketing survey (task-irrelevant
information) would significantly degrade driver performance, as compared to driving
with the human or automation navigation aiding. It was expected that there would be
higher SA scores with perfectly reliable guidance from the human advisor or automation
aid than with 80% and 60% reliable information, particularly for higher-level SA (e.g.
projection of routes).
Opposite to the above expectations for performance and SA improvements, the mobile
phone conversation with the human aid was expected to distract driver attention from the
driving environment and to degrade lower-level SA and performance. However, the
accessibility and persistence of visual information in the automated aid condition was
expected to support lower-level driver SA and navigation and speed control.
3. Results
3.1. Driving performance
Multivariate ANOVA (MANOVA) results revealed a significant effect (F(6,106)¼ 3.99,
p5 0.05) of the navigation aid reliability on the collection of performance measures.
There was no significant main effect of the navigation aid type and no interaction effect of
aid type and reliability level.
Figure 3. Example navigation instructions by human aid and automation aid display.
Situation awareness and driving performance 1357
Dow
nloa
ded
By:
[dbk
aber
@eo
s.nc
su.e
du] A
t: 19
:30
11 J
une
2007
ANOVA results revealed a significant effect of aid reliability on driving errors
(F(3,79)¼ 5.00, p5 0.01). Figure 4 presents the mean driving errors across the various
reliability conditions. (It is important to note that in many trials, participants committed
no navigation errors and, therefore, the condition means were less than 1.) In agreement
with the hypothesis, Tukey’s test revealed significantly lower errors (p5 0.05) when
participants received 100% reliable aiding, as compared to all other conditions.
ANOVA results revealed a significant effect of navigation aid reliability on variations
in driver speed control (F(3,79)¼ 3.49, p¼ 0.0217). Figure 5 presents the root mean
square error of speed control for the settings of reliability. Tukey’s test revealed
significantly greater deviations in speed (p5 0.05) when participants performed under the
control condition (i.e. telemarketing survey), as compared to the 80% and 60% reliable
aiding conditions.
3.2. Driver situation awareness
MANOVA results revealed a significant effect (F(12,135)¼ 5.01, p5 0.01) of the
navigation aid reliability on SA measures. There was no significant main effect of the
Figure 4. Driving errors across various navigation aid reliability conditions.
Figure 5. Root mean square error (RMSE) of speed control across various navigation aid
reliability conditions.
1358 R. Ma and D. B. Kaber
Dow
nloa
ded
By:
[dbk
aber
@eo
s.nc
su.e
du] A
t: 19
:30
11 J
une
2007
navigation aid type and no interaction effect. Figure 6 presents the mean Level 1, Level 2,
Level 3 and overall SA scores for the various navigation aid reliability conditions. As was
expected, the plot reveals that, on average, drivers exhibited better overall SA when aid
information was 100% reliable and the worst SA when participants were required to
address the telemarketing survey. Exceptions included Level 1 SA for the 100% reliable
condition and Level 2 SA for the 60% reliable condition. (This is addressed below.)
ANOVA results revealed a significant effect of navigation aid reliability (F(3,79)¼7.19, p5 0.01) on the percentage of correct responses to Level 1 SA queries. As was
hypothesized, Tukey’s tests revealed significantly higher (p5 0.05) perceptual knowledge
of the driving environment with the 60% and 80% reliable aiding compared to the
control condition; however, the worst perceptual knowledge occurred with the 100%
reliable aiding condition.
ANOVA results also revealed a significant effect of navigation aid reliability
(F(3,79)¼ 16.78, p5 0.01) on Level 2 SA. According to Tukey’s tests, the 100% reliable
navigation aiding produced significantly greater (p5 0.05) percent correct responses to
level 2 SA queries in comparison to all other conditions. The 80% navigation aid
reliability also produced significantly greater (p5 0.05) Level 2 SA compared to the
control condition and the 60% aid reliability condition.
Finally, ANOVA results also revealed a significant effect of navigation aid reliability
(F(3,79)¼ 3.17, p5 0.05) on Level 3 SA (projection of states of the driving environment).
Tukey’s test revealed significantly higher (p5 0.05) Level 3 SA for all navigation aid
reliability levels, as compared to the control condition. Any navigation aiding appeared
to support route planning and strategic behaviour better than none.
3.3. Correlation analyses
Simple correlation analyses were also conducted to identify any significant relationships
among driving performance and SA. A Pearson coefficient revealed a significant negative
linear association between Level 3 SA scores and driving navigation errors (r¼70.25,
p5 0.05) (i.e. as Level 3 SA increased, errors decreased). This finding was in line with
Matthews et al. (2001) theoretical linkage of Level 3 SA with strategic driving behaviour.
Figure 6. Mean percentage of correct responses to situation awareness (SA) queries for
navigation aid reliability conditions.
Situation awareness and driving performance 1359
Dow
nloa
ded
By:
[dbk
aber
@eo
s.nc
su.e
du] A
t: 19
:30
11 J
une
2007
4. Discussion
The automation and human aids appeared to cause comparable levels of distraction on
driver navigation (i.e. there were no significant differences among aid types in terms of
performance). This observation was counter to the findings of Dzindolet et al. (2002);
however, their focus was primarily on human trust in automated aids and not resulting
SA. The automation aid in this study demanded driver visual attention (Ho et al. 2006) for
perception of navigation information. The human aid may have required driver working
memory for storing verbal stimuli (the route instructions and speed limits). Driver
distraction due to mobile phone use and visual attention allocation to the laptop display
also appeared to have comparable influences on driver ability in terms of achieving SA.
Again, there were no significant differences in SA among the aid types; however, SA
accuracy across all reliability levels was far from perfect (32% at the worst and 74% at the
best). Since driving is a visual and motor control process, it is possible that the visual
search demands associated with retrieving information from the automation-aid display
also degraded driver SA and performance comparable to the impact of human use of the
mobile phone. Relevant to this, some other research has provided evidence that mobile
phone use, while driving, significantly degrades SA, particularly Levels 2 and 3 (Gugerty
et al. 2003, Ma 2006, Ma and Kaber 2005). The study by Gugerty et al. found that SA was
significantly degraded when performing a driving task while talking with a partner as
compared to only driving a car. SA was assessed in the study by using location-recall
probes requiring participants to indicate the locations of surrounding traffic.
Like Wiegmann et al. (2001), this study found that human performance in the driving
domain was generally sensitive to different levels of automation aid reliability. The
systematic decrease in the reliability of in-vehicle navigation aiding appeared to subtract
from driver attention to the roadway. This had a negative influence on driver SA and,
possibly, driving performance in terms of strategic behaviours. Driver Level 1 SA was
particularly low under the perfect aiding condition. It is possible that drivers realized the
human advisor or automated aid was highly accurate in directions under this condition
and, as a result, they did not pay as much attention to observing aspects of the roadway.
This potential behaviour was also evidenced in comparison to the 80% and 60% reliable
aiding results. Participants appeared to pay close attention to the driving environment
when they were provided with imperfect aiding. In general, these results revealed that
higher aid reliability produced higher driver SA (Levels 2, 3 and overall SA). For example,
under perfect aiding, of the information in the driving environment that participants
perceived, their relation of this information to driving goals was relatively accurate (74%).
Most importantly, this study provided empirical evidence of linkages of specific
(operational and strategic) driving behaviours to the three levels of SA. From the
experiment and correlation results, it appears that driver SA was significantly affected by
aid reliability, and driver performance was significantly correlated with changes in high-
level SA. That is, the characteristics of in-vehicle navigation aids have a mediating effect
on driver SA, which appears relevant to types of driving behaviour. It is important to
note that the results summarized here apply to idealized driving situations with limited
interactive traffic (no on-coming or overtaking vehicles).The correlation analyses for the
current study, and from Ma and Kaber (2005), revealed positive associations between SA
with one or more dimensions of driving performance. These additional findings
demonstrate the importance of the cognitive construct to driving. Additional research
is needed to further indicate exactly how each mediating factor (i.e. form of in-vehicle
automation) influences driver SA. For example, these findings suggest that perfect
1360 R. Ma and D. B. Kaber
Dow
nloa
ded
By:
[dbk
aber
@eo
s.nc
su.e
du] A
t: 19
:30
11 J
une
2007
navigation aid reliability could negatively influence Level 1 SA in navigation driving tasks
by motivating driver complacency with regard to perceiving changing states of the
roadway environment. Of course, it would not be desirable to present incorrect
navigation information to drivers in real-world systems, but methods for ensuring driver
attention to the roadway with reliable aids should be incorporated in navigation aid
design, such as periodic text or verbal message reminders.
5. Conclusions
The effect of in-vehicle navigation aid use and reliability in driving tasks on driver SA and
performance has not been previously assessed. This experimental study provided some
explanation of the mediating influence of in-vehicle automation (e.g. ACC and navigation
aids) and devices (e.g. mobile phones) on driving task performance. This said, the results
of the study should be applied with caution for making decisions about the design or use
of in-vehicle automation. The experiment was conducted with a ‘home-grown’ driving
simulation, including a fixed-base driving platform providing no kinesthetic motion cues
and there was no interactive traffic represented in the simulation runs. Drivers may (and
likely do) behave differently in actual operational settings because of the serious
consequences of having an accident. Related to this limitation, there is a need to use
highly realistic and complex driving tasks to further investigate the relationship of driver
SA with performance as a basis for future systems design. Another caveat of the study is
related to the adaptation of the SAGAT measure. Based on the training session, all
participants knew SA queries were to occur at the end of the navigation trials and
they may have taken advantage of this knowledge to prepare for quizzes. The experiment
also presented the aiding reliability conditions in the order 100%, 80% and 60%,
followed by the control condition (telemarketing survey). Although participants were
thoroughly trained on the virtual driving environment, this systematic manipulation of
aid reliability may have led to a trial order effect for the dependent variables.
Unfortunately, there were insufficient degrees of freedom provided by the datasets to
make statistical assessment of this potential issue with a full ANOVA model in aid type
and aid reliability.
The authors believe there remains a need to provide empirical evidence of the role of
driver SA in tactical driving behaviour (e.g. changing lanes, overtaking, negotiating
intersections, etc.) and to identify other mediating individual, technological and driving
system factors. Unfortunately, other recent investigations of the effects of in-vehicle
automation (e.g. ACC) on driver SA, such as the study by Stanton and Young (2005),
have not been able to confirm potential relationships between levels of driver SA
(perception, comprehension and projection) and tactical driver behaviour/action, because
they have used indirect (performance-based) measures of SA, or subjective rating
techniques, which do not assess each level of the theoretical construct, as the SAGAT
measure does. Finally, from a design perspective, the results of the present experiment
provide support for the use of in-vehicle navigation aiding under normal driving
conditions for facilitating driver SA. It is possible that the use of in-vehicle automation in
complex driving tasks, involving unexpected or hazardous conditions, may have different
effects on driver performance and SA. Additional mental resources may be required of
drivers under non-normal or hazardous conditions. This factor and other individual
variables, such as age and perceptuo-cognitive abilities, might lead to a different set of
results. Future work should investigate the effects of navigation aids under non-normal
or critical/hazardous driving conditions on driver SA and performance.
Situation awareness and driving performance 1361
Dow
nloa
ded
By:
[dbk
aber
@eo
s.nc
su.e
du] A
t: 19
:30
11 J
une
2007
Acknowledgements
The first author worked as a research assistant in the Edward P. Fitts Department of
Industrial & Systems Engineering at North Carolina State University during the
completion of this research. The authors would like to thank Xuezhong Wang for his
assistance in building all the driving navigation simulation models. We thank Sang-Hwan
Kim, who assisted with the experimental data collection.
References
DZINDOLET, M.T., PIERCE, L.G., BECK, H.P. and DAWE, L.A., 2002, The perceived utility of human and
automated aids in a visual detection task. Human Factors, 44, 79–94.
DZINDOLET, M.T., PIERCE, L.G., BECK, H.P., DAWE, L.A. and ANDERSON, B.W., 2001, Predicting misuse and
disuse of combat identification systems. Military Psychology, 13, 147–164.
ENDSLEY, M.R., 1995a, Toward a theory of situation awareness in dynamic systems. Human Factors, 37, 32–64.
ENDSLEY, M.R., 1995b, Measurement of situation awareness in dynamic systems. Human Factors, 37, 65–84.
GUGERTY, L.J., RANDO, C., RAKAUSKAS, M., BROOKS, J. and OLSON, H., 2003, Differences in remote versus in-
person communications while performing a driving task. Ergonomics Society of Kora: Seoul. In Proceedings
of XVth Triennial Congress of the International Ergonomics Association Conference (CD-ROM), 24–29 August
2003, Seoul, Korea.
HO, C., HONG, Z., TAN, H.Z. and SPENCE, C., 2006, The differential effect of vibrotactile and auditory cues on
visual spatial attention. Ergonomics, 49, 724–738.
KABER, D.B., RILEY, J.M., LAMPTON, D. and ENDSLEY, M.R., 2005, Measuring situation awareness in a virtual
urban environment for dismounted infantry training. In Proceedings of the 11th International Conference on
Human-computer Interaction, vol. 9 (Mahwah, NJ: Lawrence Erlbaum & Assoc).
MA, R., 2006, The effect of in-vehicle automation and reliability on driver situation awareness and trust.
ProQuest UMI.
MA, R. and KABER, D.B., 2005, Situation awareness and workload in driving while using adaptive cruise control
and a cell phone. International Journal of Industrial Ergonomics, 35, 939–953.
MATTHEWS, M.L., BRYANT, D.J., WEBB, R.D. and HARBLUK, J.L., 2001, Model for situation awareness and
driving. Transportation Research Record, 1779, 26–32.
PARASURAMAN, P. and RILEY, V., 1997, Humans and automation: use, misuse, disuse, abuse. Human Factors, 39,
230–253.
STANTON, N.A. and YOUNG, M.S., 2005, Driver behavior with adaptive cruise control. Ergonomics, 48, 1294–
1313.
WALKER, G.H., STANTON, N.A. and YOUNG, M.S., 2006, The ironies of vehicle feedback in car design.
Ergonomics, 49, 161–179.
WARD, N.J., 2000, Automation of task processed: an example of intelligent transportation systems. Human
Factors and Ergonomics in Manufacturing, 10, 395–408.
WIEGMANN, D.A., RICH, A. and ZHANG, H., 2001, Automated diagnostic aids: The effects of aid reliability on
user’s trust and reliance. Theoretical Issues in Ergonomics Science, 2, 352–367.
Appendix 1. An example set of situation awareness queries posed to drivers
at the close of navigation
1. What was the last road sign you saw? (L1)
Pedestrian crossing
Deer crossing
Railroad
Speed limit
Slow sign
Stop sign
I don’t know
1362 R. Ma and D. B. Kaber
Dow
nloa
ded
By:
[dbk
aber
@eo
s.nc
su.e
du] A
t: 19
:30
11 J
une
2007
2. What was the colour of the vehicle behind your car when the simulation stopped?
(L1)
No car
Grey
White
I don’t know
3. What was your vehicle speed (mph) at the time the simulation stopped? (L1)
Less than 25
25–30
30–35
35–40
40–45
45–50
More than 50
4. How long has it been since you passed the last turn in navigating the city suburb? (L2)
Less than 30 s
30 s–1 min
1–1.5 min
1.5–2 min
2–2.5 min
2.5–3 min
More than 3 min
5. How long has it been since you passed the last road sign? (L2)
1–5 s
5–10 s
10–15 s
15–20 s
20–25 s
More than 25 s
6. In which direction from your vehicle was your destination (building) located when
you passed the last turn in the city suburb? (L2)
On the left
On the right
Right in front of me
Behind me
I don’t know
7. How long did you think it would take for your vehicle to reach the destination in the
simulation after you passed the last road sign? (L3)
No sign in sight
1–5 s
5–10 s
10–15 s
15–20 s
More than 20 s
Situation awareness and driving performance 1363
Dow
nloa
ded
By:
[dbk
aber
@eo
s.nc
su.e
du] A
t: 19
:30
11 J
une
2007
8. What was the optimal navigation route to reach your destination when you passed
the intersection of Kaber St. and Ma St. (what route would have generated the
shortest drive time)? (L3)
Go to Ma St., then Riley Rd, then McDowell St
Go to Kaber St., then Noa Dr., then Riley Rd., then McDowell St.
Go to Ma St., then Avent Ferry St.
I don’t know
9. How long would the time have been from the intersection of Kaber St. and Ma St.
until you finished driving through the suburban area on the ‘optimal’ route? (L3)
Less than 1 min
1–2 min
2–3 min
3–4 min
4–5 min
More than 5 min
1364 R. Ma and D. B. Kaber