Content uploaded by Anna Feldhütter
Author content
All content in this area was uploaded by Anna Feldhütter on Nov 19, 2018
Content may be subject to copyright.
This manuscript is a post-print of Feldhütter, A., Kroll, D., & Bengler, K. (2018). Wake up and take over! The effect of fa-
tigue on the take-over performance in conditionally automated driving. In IEEE ITSC 2018: 21th International Confer-
ence on Intelligent Transportation Systems: Maui, Hawaii, USA, November 04-07, 2018 (pp. 2080–2085).
978-1-7281-0322-8
Wake Up and Take Over! The Effect of Fatigue on
the Take-over Performance in Conditionally
Automated Driving
Anna Feldhütter
Chair of Ergonomics
Technical University of Munich
Garching, Germany
anna.feldhuetter@tum.de
Dominik Kroll
Chair of Ergonomics
Technical University of Munich
Garching, Germany
dominik.kroll@tum.de
Klaus Bengler
Chair of Ergonomics
Technical University of Munich
Garching, Germany
bengler@tum.de
Abstract— Although fatigue’s negative impact on driving
performance is well known from manual driving, its effect on
the take-over performance during the transition from condi-
tionally automated driving to manual driving is still uncertain.
The effect of fatigue on the take-over performance was exam-
ined in a driving simulator study with 47 participants assigned
to two conditions: fatigued or alert. In the corresponding con-
dition (fatigued or alert), the desired driver state was promot-
ed by specific measures (e.g. daytime, caffeinated beverages,
physical exercise). In the fatigued condition, the take-over situ-
ation was triggered once participants reached a certain high
level of fatigue. Two trained, independent observer assessed
fatigue with the support of a technical fatigue assessment sys-
tem based on objective eyelid-closure metrics (e.g. PERCLOS).
In the alert condition, participants drove conditionally auto-
mated for a fixed 5-minute period. Results showed no signifi-
cant difference between participants’ take-over times in the
two conditions. However, fatigued participants were signifi-
cantly more burdened and stressed during the take-over situa-
tion than were alert participants and manifested less confident
behavior when coping with the situation. This behavior may
negatively affect the transition from conditionally automated
driving to manual driving in more complex situations and mer-
its further examination.
Keywords—fatigue, conditionally automated driving, take-
over performance
I. INTRODUCTION
A. Driver Fatigue in Road Traffic
Many past international studies have identified driver fa-
tigue as one of the major factors contributing to accidents in
road traffic [1–3]. Depending on the study, up to 25–31% of
accidents are caused mainly by fatigue [4–6]. The various
effects of fatigue on the driving performance might explain
these high rates. Fatigue may not only diminish drivers’ at-
tention to the traffic environment [7] and their response
(time) to unusual events or emergency situations [5, 8–10],
but also affect steering control [11] and speed maintenance
[12]. According to the model presented in [13], which ex-
tends the findings documented in [14], driver fatigue can be
distinguished into task-related (TR) and sleep-related (SR)
fatigue based on the underlying cause. SR fatigue results
from sleep deprivation or disturbance of the circadian
rhythm, whereas TR fatigue occurs due to the driving task
and driving environment, which either impose high cognitive
demands over a prolonged time interval (causing active TR
fatigue) or produce mental underload (causing passive TR
fatigue). Dense traffic or poor visibility, for instance, result
in high task demand. Mental underload can be produced by
largely automated systems and the resulting monotony. Re-
gardless of the causal factors, fatigue leads to the stated per-
formance deterioration.
Fully automated vehicles, which no longer require human
input and treat the driver as a passenger, promise to eliminate
accidents caused by fatigue in road traffic. However, at low-
er automation levels of the SAE taxonomy [15], such as con-
ditional automation (CA), the driver is still considered to be
the fallback when system limits are reached. This means that
although the driver is allowed to engage in non-driving-
related activities (NDRA) a timely and proper reaction to a
take-over request (TOR) is required. Therefore with CA, the
driver still bears safety-critical responsibilities making ade-
quate take-over performance essential. Whether or not fa-
tigue impairs take-over performance is still in question.
B. Driver Fatigue in Automated Driving
Past research has yielded controversial findings concern-
ing the effect of fatigue on the take-over performance. Saxby
et al. [10] found that passive fatigue evoked by prolonged
phases of automated driving (10 and 30 minutes) caused a
deterioration of response times to an emergency event and an
increased crash probability. The findings reported in [16]
also revealed impaired take-over performance in that way
that participants showed a decrement of lateral control in the
take-over situation. By contrast, [17] who compared the
take-over performance after 5 and 20 minutes of conditional-
ly automated driving (CAD) found no impairment. Feldhüt-
ter [18] compared the take-over behavior of drivers fatigued
by a prolonged duration of CAD without an activity to that
of drivers who were aroused by engagement in voluntary
NDRA during 60 minutes of CAD. The take-over perfor-
mance showed no impairment. On the contrary, participants
in the NDRA group demonstrated significantly stronger
braking action during the take-over situation suggesting that
they had less situation awareness. Weinbeer [19] simulated a
TOR using a response-time task in a Wizard-of-Oz study
where participants had to input a steering reaction after the
take-over sound was triggered. Results showed no deteriora-
tion in the response times.
The discrepancies among these studies’ findings might
be attributable to widely divergent methods for assessing
fatigue and significantly different study designs. Saxby [10]
and Feldhütter [17] triggered the TOR after a specific dura-
tion of automated driving regardless of the prevailing fatigue
state. Goncalves [16] and Weinbeer [19] triggered a take-
over situation once a specific fatigue level was reached.
However, assessment of fatigue level required to trigger a
TOR differed significantly. Whereas [16] used self-rating of
fatigue on the Stanford Sleepiness Scale, [19] relied on an
observer rating based on specific fatigue indicators to assess
fatigue.
C. Purpose of this Study
We compared fatigued and alert drivers in the context of
a driving simulator experiment to contribute to greater trans-
parency in terms of the effect of fatigue on the take-over
performance. We chose a fatigue-state-dependent study de-
sign resembling those used in [19] and [16]. To assess fa-
tigue and identify the point in time when the desired fatigue
level was reached, we decided to use a method combining
the fatigue assessment system proposed in [20] with human
observer rating.
II. METHODS
A. Participants
A total of 57 participants took part in the experiment.
Technical problems caused three participants to be excluded
from data analysis. Seven (23%) of the 30 participants tested
in the fatigued condition did not reach the defined fatigue
level within the predefined maximum time interval and were
also excluded from the analysis. The remaining sample con-
sisted of 47 participants (22 fatigued, 25 alert) with 11 fe-
males (23%). Mean age was 24.32 years (SD
=
4.01 years)
ranging from 21 to 47 years. Criterion for participation was
the possession of a valid driver’s license for at least four
years to prevent inexperienced drivers’ behavior from bias-
ing the results.
B. Study Design and Measures
To examine the effect of fatigue on take-over perfor-
mance, we chose a between-subjects design consisting of
two conditions (fatigued and alert). Participants were indi-
vidually tested and each was assigned to one of the two con-
ditions prior to the experiment. Participants in neither condi-
tion were offered any NDRA during the test drive to avoid
the influencing factor distraction on the driver state. Partici-
pants of the fatigued condition drove conditionally automat-
ed at a constant velocity of 120
km/h (74.57
mph) in the right
lane of a three-lane freeway until the defined fatigue state
was reached, but for feasibility reasons no longer than
90 minutes (see Fig. 1). This was assumed to be reasonable
since [18] revealed that the maximum fatigue level of all
participants was reached between 15 and 45 minutes of CAD
and [19] found no further increase in fatigue after 75 minutes
of CAD. The test track was designed to be as monotonous as
possible meaning that the road was uniformly straight, the
surrounding traffic unvaried and sparse, and no diversifying
elements such as different landscape features were imple-
mented. To further promote the development of fatigue, par-
ticipants were instructed in advance not to drink caffeinated
beverages on the day of the experiment. Further, we started
the experiment before 8 a.m., after lunch between 1 p.m. and
2 p.m. or after 6 p.m.
Participants in the alert condition were invited between
10 a.m. and 12 p.m. or between 3 p.m. and 5 p.m. They
drove a fixed period of 5 minutes conditionally automated
(see Fig. 1). The simulated track was designed diversely with
curves, an interesting landscape, and maneuvers such as
speed changes and overtaking to keep participants alert. Fur-
thermore, participants had to engage in light physical exer-
cise (1 minute of rope jumping) right before starting the ex-
periment to further promote alertness.
After each group’s (alert and fatigued) designated auto-
mation duration, all participants experienced the same take-
over situation initiated by a broken-down vehicle suddenly
appearing in the lane of the ego-vehicle (simulated vehicle of
the participant) representing a system limit. The take-over
was requested by an acoustic alert in form of a doubled beep
and a notice on the instrument panel. The obstacle appeared
200 meters away, so the time budget for taking over control
and resolving the situation by braking or steering was
6 seconds. During the TOR, the ego-vehicle was located in
the right lane while two other vehicles were located in the
middle lane respectively 50 meters ahead and behind the
ego-vehicle. The left lane was free of traffic.
1) Assessing Fatigue
A valid method for assessing fatigue is required to realize
a fatigue-state-dependent study design. For this, two trained
observers independently rated participants’ fatigue during
the test drive according to the scale documented in [21]. This
scale was adapted from the common Wiewille-scale [22] and
includes specific behavioral indicators for fatigue (see Table
1). Besides, the observers were supported by an objective
fatigue-assessment system (FAS) proposed in [20], which
was implemented in the driving simulator. The system fuses
multiple eye-tracking metrics (PERCLOS, number of micro-
sleep events and head movements) and classifies the driver
as fatigued or alert. PERCLOS indicates the percentage of
time within 60 seconds during which the eyes are more than
80% closed, which is a strong indicator of fatigue [23]. Both
observers rating a participant at fatigue level 3 or 4 and the
FAS simultaneously classifying her or him as fatigued trig-
gered a take-over situation. In cases of insufficient FAS reli-
ability (mostly due to poor data quality), rater classification
alone also sufficed to trigger a take-over situation.
60
5
0
TOR
30
Duration of CAD (minutes)
Fatigued
condition
90
TOR
Fixed
Variable (fatigue-state-dependent)
Alert
condition
Fig. 1. Study design
TABLE I. LIST OF FATIGUE INDICATORS USED FOR OBSERVER
RATING ACCORDING TO [21] AND ADAPTED FROM [22]
Fatigue
level
Fatigue indicators
1 (not
fatigued)
Normal fast eye blinks (<0.5
s), short ordinary glances
(fast saccades), normal facial tone, occasional body
movements or gestures
2 (fatigued)
Slower eyelid closures (0.5
s to 1
s), rare saccades and
prolonged fixations (glassy-eyed appearance), more
frequent blinks, decreased facial tone (“flaccid face”),
behavior to counteract fatigue: mannerisms (rubbing
the face or eyes, scratching, facial contortions), moving
restlessly in the seat
3 (very
fatigued)
Long eyelid closures (1
s to 2
s), rolling upward or a
sideways movement of the eyes, lacking fixations, rare
blinks, further decreased facial tone, lack of apparent
activity and large, isolated (or punctuating) move-
ments, such as providing a large correction to steering
or reorienting the head from a leaning or tilting posi-
tion.
4 (extremely
fatigued)
Occurrence of microsleep: eyelid closures of >2
s, lack
of movements, at the transition to microsleep: head
nodding, jerky movements, raising head or body
2) Assessing Take-over Performance
The take-over performance was operationalized by vari-
ous dependent variables. The take-over time (ToT) is the
time participants need for the first conscious intervention
maneuver as a reaction to the TOR, either by steering (steer-
ing wheel angle > 2
°), breaking, or accelerating (change of
braking pedal or gas pedal position > 10
%) [24]. Further, the
maximum longitudinal (AccLong) and lateral acceleration
(AccLat), and the minimum time-to-collision (TTC) were
analyzed. The TTC is the time theoretically remaining until a
potential collision with an obstacle assuming constant speed
of the ego-vehicle. Greater accelerations and smaller TTCs
indicate worse take-over quality [25]. Furthermore, we ana-
lyzed the maneuver participants ultimately chose to resolve
the take-over situation. Maneuver options were full braking
in the ego-lane or changing the lane to avoid the obstacle.
Rearward or sideward traffic needs to be checked to deter-
mine whether it allows the corresponding maneuver. Thus,
we analyzed whether participants consciously looked into the
rearview or side-view mirror for a lane changing maneuver
or the rearview mirror for full braking maneuver after the
TOR. We termed the metric visual securing behavior
(SecB). Changing lane without checking the environment is
more safety-relevant than full braking without checking.
C. Apparatus
1) Driving Simulator and Automation
The driving experiment took place in the high-fidelity,
fix-based driving simulator of the Chair of Ergonomics. The
simulator consisted of a complete BMW E64 mockup. Six
projectors created a 180° field of view allowing the use of
the side-view and the rearview mirrors. An in-vehicle audio
system generated engine sounds and road noise. SILAB
software from the Wuerzburg Institute for Traffic Sciences
(WIVW GmbH) was used for the driving simulation. Driv-
ing-related data was sampled at 60
Hz.
The implemented CA took over longitudinal and lateral
control of the vehicle including passing maneuvers whenever
necessary. Participants had to press a button on the active
steering wheel to activate and deactivate CA. Manually
steering or braking above a specific threshold would also
deactivate the automation.
2) Camera Systems
The eyes and the head of the participants were tracked
and recorded with a frequency of 60 Hz by the three-camera,
remote eye-tracking system SmartEye. The real-time video
film that the camera system delivered showing participants’
faces and upper torsos also provided the basis for the online
observer rating of participants’ fatigue.
D. Procedure
The experiment was conducted in German. Before the
experiment, participants were asked to fill in a questionnaire
containing demographic questions and received a specific
introduction depending on the experimental condition. After
welcoming the participants, the examiner gave an introduc-
tion to the experiment and the participants could become
familiar with the driving simulator and the automation sys-
tem during a training drive of about 7 minutes. Before enter-
ing the driving simulator, all participants were asked to leave
watches, smartphones or other personal items outside to pre-
vent an awakening effect by a NDRA. Participants of the
alert condition subsequently completed the brief exercise.
The eye-tracking system was calibrated after participants
entered the mockup. Afterwards, participants experienced
the condition-specific experimental track. The post-interview
was conducted right after the take-over situation was re-
solved and the simulation was stopped. Participants had to
rate their fatigue level retrospectively right before the TOR
on the nine-level Karolinska Sleepiness Scale (KSS) [26]
(1 =
“Extremely alert”; 9
=
“Very sleepy, great effort to keep
awake, fighting sleep”). Furthermore, participants were
asked how many hours they slept the night before the exper-
iment and how they perceived the criticality of the take-over
situation on a 10-level scale ranging from 1
=
“Not critical”
to 10
=
“Extremely critical”.
III. RESULTS
A two-tailed t-test was performed to analyze the effect of
fatigue on the take-over performance. Welch’s t-test was
used in the case of a significant Levene’s test (assumption of
equal variances is violated). The Chi-square test was con-
ducted to analyze binominal data. For frequencies ≤ 5, Fish-
er’s exact test was used instead. An α
=
0.05 significance
level was set for all statistical analyses.
A. Post-interview and the Development of Fatigue
The post-interview revealed that the majority of partici-
pants in the fatigued condition slept 5–6 hours (41%) or less
than 5 hours (9%) the night before the experiment. In com-
parison, nobody of the alert condition slept less than 5 hours
the night before the experiment. Most participants in this
condition, stated that they slept 6–7 hours (40%) or 7–8
hours (20%).
On average, participants in the fatigued condition drove
42.4 minutes (SD
=
18.05
min; min
=
19
min; max
=
80
min)
conditionally automated until they reached the designated
level of fatigue and the take-over situation was triggered.
Only one participant (5%) in this condition drove longer than
75 minutes. Results of the KSS showed that 27% of the par-
ticipants in the fatigued condition retrospectively rated them-
selves at level 7 (“Sleepy, but no effort to keep awake”),
50% at level 8 (“Sleepy, some effort to keep awake”), and
23% at level 9 during the moment right before the TOR. The
online observers’ ratings of fatigue revealed that 86% of the
22 fatigued participants were rated in fatigue level 3 when
the TOR was triggered, and 14% in level 4. Also using the
PERCLOS metric and the result of observer rating during the
entire automated drive to analyze the development of fatigue
shows that, on average, fatigue increased steadily over time
(see Fig. 2). In the alert condition, 68% of the 25 participants
rated themselves at level 2 (“Very alert”) or level 3 (“Alert”)
on the KSS. Four participants (16%) stated that they were at
level 4 (“Rather alert”) and three (12%) at level 5 (“Neither
alert nor sleepy”). Only one participant (4%) quoted to be at
level 6 (“Some signs of sleepiness”).
B. Effect of Fatigue on the Take-over Performance
Fig. 3 and Fig. 4 show the take-over parameters ToT, TTC,
AccLong, and AccLat dependent on the condition (fatigued,
alert). The statistical analysis of these parameters revealed
that there was no significant difference between the alert and
the fatigued conditions for the ToT, TTC, and AccLat
whereas participants in the fatigued condition tended to pro-
duce higher TTCs (ΔM
=
0.28
s, t(43)
=
−1.69, p
=
0.098). The
Welch’s t-test revealed that participants in the fatigued con-
dition produced significant greater AccLong (ΔM
=
2.5
m/s2,
t(40.60)
=
−2.23, p
=
0.031). One participant in each condition
failed to react to the TOR in a timely manner and collided
with the broken-down vehicle.
Table 2 shows the frequency of the maneuvers conducted to
resolve the take-over situation and whether the participants
checked the mirrors or the environment (SecB) before exe-
cuting the corresponding maneuver depending on condition.
Fatigued participants (45%) decided to conduct a full brak-
ing maneuver (χ2(1, N
=
47)
=
8.63, p
=
0.003) significantly
more frequently than alert participant (8%). Only 30% of the
10 fatigued participants and none of 2 alert participants who
conducted full braking maneuvers checked the rearview mir-
ror. For the lane changing maneuver, the SecB was more
balanced: 83% of the fatigued and 87% of the alert partici-
pants checked the mirrors before changing the lane.
p<0.05
Fatigued
Alert
0.0
0.5
1.0
1.5
2.0
2.5
3.0
3.5
4.0
4.5
0.00
0.01
0.02
0.03
0.04
0.05
0.06
0.07
0.08
0.09
10 20 30 40 50 60 70 80 90 100
Level of fatigue
PERCLOS
Relative duration of the experimental drive [%]
PERCLOS
Expert rating
Acceleration [m/s2]
Time [s]
Fatigued
Alert
Fig. 2. Development of fatigue in the fatigued condition assessed
using mean PERCLOS and expert rating plotted against duration as
a percentage (%) of the overall duration of each participant’s
conditionally automated drive; error bars represent the standard
error
Fig. 3. Boxplots of take-over time (ToT)
and minimal time-to-collision (TTC)
depending on condition; x represents the
mean value
Fig. 4. Boxplots of maximum
longitudinal and lateral acceleration
(AccLong, AccLat) depending on
condition; x represents the mean
value
TABLE II. FREQUENCY OF THE TWO POSSIBLE MANEUVERS (FULL BRAKING OR LANE CHANGE), THE FREQUENCY OF VISUAL SECURING (SECB)
BEFORE CONDUCTING THE MANEUVER DEPENDING ON CONDITION
Condi-
tion
Maneuver
Full braking
Lane Change
Total
Maneuver
frequency
(absolute)
Maneuver
frequency
(%)
SecB
frequency
(absolute)
SecB
frequency
(%)
Maneuver
frequency
(absolute)
Maneuver
frequency
(%)
SecB
frequency
(absolute)
SecB
frequency
(%)
Maneuver
frequency
(absolute)
SecB
frequency
(absolute)
SecB
frequency
(%)
Fatigued
10
45
3
33
12
55
10
83
22
13
59
Alert
2
8
0
0
23
92
20
87
25
20
80
Total
12
26
3
25
35
74
30
86
47
33
70
The analysis of the criticality rating demonstrated that partic-
ipants in the fatigued condition (M
=
5.16, SD
=
2.17) per-
ceived the same situation to be significantly more critical
than did participants in the alert condition (M
=
6.86,
SD
=
2.67), t(45)
=
−2.36, p
=
0.023.
IV. DISCUSSION
Results of the post-interview indicated that manipulation
of the two conditions succeeded. All participants in the fa-
tigued condition placed themselves between levels 7 and 9
on the KSS, which is rated as sleepy, and all participants in
the alert condition placed themselves between levels 2 and 6,
which can be regarded as an active state [27]. The develop-
ment of PERCLOS and observers’ fatigue rating during the
test drive also support these results since a steady increase
appeared for both metrics. On average, it took participants 42
minutes of conditionally automated driving to reach a high
level of fatigue. The wide range of testing durations—
between 19 and 80 minutes of conditionally automated driv-
ing—substantiates the hypothesis that fatigue development is
highly individual, which is in line with previous studies [18,
28]. Furthermore, the individual testing durations confirm
the use of a fatigue-state-dependent study design.
Due to previous findings [9, 10, 29] we expected to find
greater take-over times in the fatigued condition, which
could not be confirmed in this study. However, participants
evolved a different interesting behavior: Almost half of the
fatigued participants executed a full-braking maneuver to
resolve the situation involving the broken-down vehicle.
This produced significantly greater decelerations (AccLong)
and, by tendency, higher TTCs in this condition. The majori-
ty did not check the mirrors or the environment before break-
ing. This behavior is assumed to have resulted from a startle
reaction rather than from a reasoned decision. Assuming that
fatigue diminishes attention to the environment [7], fatigued
participants lacked sufficient situation awareness when it
came to the TOR. Startled by the TOR, fatigued participants
showed short response times comparable to those of the alert
drivers. However, the complex, suddenly developing situa-
tion overwhelmed the fatigued participants to the extend that
they were unable to rationally decide how to react and re-
sponded instead with a fast overreaction. This involved full
brake application (until stopping) before doing anything else,
such as checking the environment, in order to buy time to
better evaluate the situation. By contrast, more than 90% of
the alert participants decided to change lane to resolve the
situation; the majority also checked the environment before
executing the maneuver. This behavior indicates that the
alert participants reacted consciously and deliberately. The
criticality rating also supports this assumption since partici-
pants in the fatigued condition perceived the same take-over
situation to be significantly more critical than did partici-
pants in the alert condition, suggesting that fatigued partici-
pants felt more overloaded and stressed.
In other words, fatigue might lead to an overreaction in
take-over situations caused by overload, which is relevant for
the safe operation of vehicles with CA and which has im-
portant implications for the future application. This behavior
did not lead to an increased crash probability in this study;
however, different take-over situations need to be tested to
further investigate these findings. It is conceivable that fa-
tigued drivers’ less confident behavior might lead to a more
obvious deterioration of the driving performance in even
more complex situations.
V. CONCLUSION
In this driving simulator study, we compared the
take-over performance of fatigued and alert drivers during
the transition from CAD to manual driving. Specific
measures were taken to promote fatigue or alertness depend-
ing on the condition. Fatigue was assessed using a combina-
tion of observer rating conducted by two independent human
raters and a certain technical system based on objective eye-
lid-closure metrics. We showed that the method applied in
this study to manipulate and assess fatigue has evidently
succeeded. Furthermore, we were able to conclude that fa-
tigue did not have the expected impact on the take-over per-
formance, namely prolonging take-over times and producing
more accidents in the take-over situation. However, we did
find a significant, distinct take-over behavior specific to fa-
tigued drivers. Fatigued drivers appeared to overreact in such
a way that they conducted rather an unsecured minimal-risk
maneuver for reducing the risk of collision than a conscious-
ly planned maneuver. By contrast, alert drivers mostly react-
ed with careful consideration and consequently more safely.
These findings may have a safety-relevant impact on the
future application of vehicles with CA.
REFERENCES
[1] J. Connor et al., “Driver sleepiness and risk of serious injury to car
occupants: population based case control study,” British Medical
Journal, vol. 324, no. 7346, 2002.
[2] R. Knipling and J.-S. Wang, “Crashes and fatalities related to driver
drowsiness/fatigue,” Research Note, Washington, DC, 1994.
[3] I. D. Brown, “Driver fatigue,” (eng), hum factors, vol. 36, no. 2, pp.
298–314, 1994.
[4] S. G. Klauer, T. A. Dingus, V. L. Neale, J. D. Sudweek, and D. J.
Ramsey, “The impact of driver inattention on near-crash/crash risk:
an analysis using the 100-car naturalistic driving study data,”
Technical Report DOT HS 810 594, Virginia Tech Transportation
Institute, Blacksburg, Virginia, USA, 2006.
[5] T. von Jan, T. Karnahl, K. Seifert, J. Hilgenstock, and R. Zobel,
Don’t sleep and drive—VW’s fatigue detection technology.
Washington, DC: Proceedings – 19th International Technical
Conference on the Enhanced Safety of Vehicles, 2005.
[6] X. Zhao and J. Rong, “The relationship between driver fatigue and
monotonous road environment,” in Atlantis Computational
Intelligence Systems, Computational Intelligence for Traffic and
Mobility, W. Wang and G. Wets, Eds., Paris: Atlantis Press, 2013,
pp. 19–36.
[7] Y.-C. Liu and T.-J. Wu, “Fatigued driver’s driving behavior and
cognitive task performance: Effects of road environments and road
environment changes,” Safety Science, vol. 47, no. 8, pp. 1083–1089,
2009.
[8] S. K.L. Lal and A. Craig, “Driver fatigue: Electroencephalography
and psychological assessment,” Psychophysiol., vol. 39, no. 3, pp.
313–321, 2002.
[9] D. J. Saxby et al., “Effect of active and passive fatigue on
performance using a driving simulator,” Proceedings of the Human
Factors and Ergonomics Society Annual Meeting, vol. 52, no. 21, pp.
1751–1755, 2008.
[10] D. J. Saxby, G. Matthews, J. S. Warm, E. M. Hitchcock, and C.
Neubauer, “Active and passive fatigue in simulated driving:
discriminating styles of workload regulation and their safety
impacts,” (eng), Journal of experimental psychology. Applied, vol.
19, no. 4, pp. 287–300, 2013.
[11] P. Thiffault and J. Bergeron, “Monotony of road environment and
driver fatigue: A simulator study,” Accident Analysis & Prevention,
vol. 35, no. 3, pp. 381–391, 2003.
[12] S. K.L. Lal and A. Craig, “A critical review of the psychophysiology
of driver fatigue,” Biological Psychology, vol. 55, no. 3, pp. 173–194,
2001.
[13] J. F. May and C. L. Baldwin, “Driver fatigue: The importance of
identifying causal factors of fatigue when considering detection and
countermeasure technologies,” Transportation Research Part F:
Traffic Psychology and Behaviour, vol. 12, no. 3, pp. 218–224, 2009.
[14] P. A. Desmond and P. A. Hancock, “Active and passive fatigue
states.,” in Human factors in transportation series, Stress, workload,
and fatigue, P. A. Hancock, Ed., Mahwah NJ u.a.: Erlbaum Ass,
2001, pp. 455–465.
[15] Taxonomy and Definitions for Terms Related to Driving Automation
Systems for On-Road Motor Vehicle, J 3016_201609, 2016.
[16] J. Goncalves, R. Happee, and K. Bengler, “Drowsiness in conditional
automation: Proneness, diagnosis and driving performance effects,”
in 2016 IEEE 19th International Conference on Intelligent
Transportation Systems (ITSC): Windsor Oceanico Hotel, Rio de
Janeiro, Brazil, November 1-4, 2016, Rio de Janeiro, Brazil, 2016, pp.
873–878.
[17] A. Feldhütter, C. Gold, S. Schneider, and K. Bengler, “How the
duration of automated driving influences take-over performance and
gaze behavior,” in Advances in Ergonomic Design of Systems,
Products and Processes: Proceedings of the Annual Meeting of GfA
2016, C. Schlick et al., Eds., 1st ed., Berlin: Springer, 2017, pp. 309–
318.
[18] A. Feldhütter, T. Hecht, L. Kalb, and K. Bengler, “Effect of
prolonged periods of conditionally automated driving on the
development of fatigue - with and without non-driving-related
activities,” Cognition Technology and Work, 2018. in press.
[19] V. Weinbeer et al., “Highly automated driving: How to get the driver
drowsy and how does drowsiness influence various take-over
aspects?,” 8. Tagung Fahrerassistenz, Einführung hochautomatisiertes
Fahren, 22. - 23. November 2017, München, 2018.
[20] A. Feldhütter, A. Feierle, L. Kalb, and K. Bengler, “A new approach
for a real-time non-invasive fatigue assessment system for automated
driving,” Proceedings of the Human Factors and Ergonomics Society
Annual Meeting, 2018. in press.
[21] K. Karrer-Gauß, “Prospektive Bewertung von Systemen zur
Müdigkeitserkennung: Ableitung von Gestaltungsempfehlungen zur
Vermeidung von Risikokompensation aus empirischen
Untersuchungen,” Dissertation, Technische Universität Berlin, Berlin,
2011.
[22] W. W. Wierwille and L. A. Ellsworth, “Evaluation of driver
drowsiness by trained raters,” (eng), Accident; analysis and
prevention, vol. 26, no. 5, pp. 571–581, 1994.
[23] W. Wierwille, L. Ellsworth, S. Wreggit, R. Fairbanks, and C. Kirn,
“Research on vehicle-based driver status/performance monitoring:
Development, validation, and refinement of algorithms for detection
of driver drowsiness,” DOT HS 808 247, HighwayTraffic Safety
Administration, 1994.
[24] C. Gold, D. Damböck, L. Lorenz, and K. Bengler, “"Take over!":
How long does it take to get the driver back into the loop?,”
Proceedings of the Human Factors and Ergonomics Society Annual
Meeting, vol. 57, no. 1, pp. 1938–1942, 2013.
[25] C. Gold and K. Bengler, “Taking over control from highly automated
vehicles,” in Proceedings of the 5th International Conference on
Applied Human Factors and Ergonomics AHFE 2014, Ahram T., W.
Karwowski, and T. Marek, Eds., Kraków, Poland, 2014, pp. 3662–
3667
[26] T. Åkerstedt and M. Gillberg, “Subjective and objective sleepiness in
the active individual,” Int J Neurosci, vol. 52, no. 1-2, pp. 29–37,
1990.
[27] L. Pauly and D. Sankar, “Detection of drowsiness based on HOG
features and SVM classifiers,” in 2015 IEEE International
Conference on Research in Computational Intelligence and
Communication Networks (ICRCICN), Kolkata, India, 2015, pp.
181–186.
[28] N. Schömig, V. Hargutt, A. Neukum, I. Petermann-Stock, and I.
Othersen, “The Interaction between highly automated driving and the
development of drowsiness,” Procedia Manufacturing, vol. 3, pp.
6652–6659, 2015.
[29] D. J. Mascord and R. A. Heath, “Behavioral and physiological indices
of fatigue in a visual tracking task,” Journal of Safety Research,
vol. 23, no. 1, pp. 19–25, 1992.