Hand-over situations in highly automated driving occur when
drivers have to take over vehicle control due to automation
shortcomings. Due to high visual processing demand of the
driving task and time limitation of a takeover maneuver, appropriate
user interface designs for take over requests (TOR)
are needed. In this paper, we propose applying ambient TORs,
which address the peripheral vision of a driver. Conducting
an experiment in a driving simulator, we tested a) ambient
displays as TORs, b) whether contextual information could be
conveyed through ambient TORs, and c) if the presentation
pattern of the contextual TORs has an effect on takeover behavior.
Results showed that conveying contextual information
through ambient displays led to shorter reaction times and
longer times to collision without increasing the workload. The
presentation pattern however, did not have an effect on take
over performance.
5. Levels of Automation (NHTSA)
6
Level 0: No vehicle autonomy, driver has control
Level 1: vehicle provides driver info/warnings, driver has informed control
Level 2: vehicle integrates detection/response, driver ready to take
control
Level 3: vehicle fully autonomous, driver takes control in emergency
Level 4: vehicle fully autonomous, occupants do not need ability to drive
Sadeghianborojeni et al., Auto-UI 2016, Ann Arbor, MI, USA04.11.2016
6. Human vs Machine
Battle of the sensors
7Sadeghianborojeni et al., Auto-UI 2016, Ann Arbor, MI, USA04.11.2016
8. vigilance: monitor for unexpected events
concentrate: driving, monitor busy traffic
switching: different information locations
share: other tasks
suppress: inhibit unnecessary actions
preparation: initiate action procedures
goal-setting: maintenance of objective(s)
Software of Attention?
e.g. on the highway looking for an exit
9
Stuss, Shallice, Alexander, & Picton (1995). A multi-disciplinary approach to anterior attentional functions. In Grafman, Holyoak, Boller (Eds.
Structure and function of the human prefrontal cortex, Annals of New York Academy of Sciences, 279, 191--211
Sadeghianborojeni et al., Auto-UI 2016, Ann Arbor, MI, USA
9. vigilance: right lateral mid-frontal regions
concentrate: anterior cingulate
switching: dorsolateral prefrontal cortex
share: orbitofrontal and anterior cingulate
suppress: bilateral orbitofrontal areas
preparation: pre-motor cortex
goal-setting: dorsolateral prefrontal cortex
Software of Attention?
e.g. on the highway looking for an exit
10
Stuss, Shallice, Alexander, & Picton (1995). A multi-disciplinary approach to anterior attentional functions. In Grafman, Holyoak, Boller (Eds.
Structure and function of the human prefrontal cortex, Annals of New York Academy of Sciences, 279, 191--211
Sadeghianborojeni et al., Auto-UI 2016, Ann Arbor, MI, USA
10. Assisting Takeover Situations
11
How can we support a driver’s ability to switch from
engaging with a non-vehicle-handling task to
monitor and/or resume the complex maneuvers
that constitute effective vehicle handling?
Attention disengagement
from non-driving task
Shifting attention to
manual driving task
switching: different information locations
preparation: initiate action procedures
suppress: inhibit unnecessary actions
Sadeghianborojeni et al., Auto-UI 2016, Ann Arbor, MI, USA04.11.2016
12. Audio-Visual Take-over Request
Goal
• Shift attention from the secondary task to the driving task
• Communicate the driving environment and upcoming task
• Prepare appropriate maneuver
Attention disengagement
from non-driving task
Shifting attention to
manual driving task
Shift driver‘s attention
Shift driver‘s attention
and provide context
13. Design: Take-over Requests
14
• Ambient light displays can be effective to prime a
take-over situation.
• Presenting contextual information as TORs to
drivers, affects their performance
• The presentation pattern of the light cues affects
drivers’ performance
Sadeghianborojeni et al., Auto-UI 2016, Ann Arbor, MI, USA04.11.2016
21. Measurements
22
Reaction time (RT)
the time between presentation of the TOR and first
steering action
Time to collision (TTC) to obstacle
the time between the lane change maneuver and
collision to the road block
Workload (NASA-RTLX)
self-reported workload ratings
Gaze behavior
number and duration of glances at the light display
when the TORs were presented
Sadeghianborojeni et al., Auto-UI 2016, Ann Arbor, MI, USA04.11.2016
24. Qualitative Feedback
• “Having light in the periphery together with the
auditory cue, attracts attention faster to the handover
task. ”
• “ It saves time scanning the road and seeing what is
wrong and what I have to do.”
• 85% of the participants preferred conditions with
contextual cuing (static and moving) to the baseline.
• Between the static and moving lights, the moving light
was preferred (71%)
25Sadeghianborojeni et al., Auto-UI 2016, Ann Arbor, MI, USA04.11.2016
25. Reaction Times (msecs)
F2,38 = 7.46, p < 0.01, ω2 = 0.24
Bayes Factor ≈ p(H0):p(H1) =3.58
26
*
*
Baseline Static Moving
Sadeghianborojeni et al., Auto-UI 2016, Ann Arbor, MI, USA04.11.2016
Kass, R. E., & Raftery, A. E. (1995). Bayes factors. Journal of the american statistical association, 90(430), 773-795.
26. Reaction Times (msecs)
F2,38 = 7.46, p < 0.01, ω2 = 0.24
H0 is 3.58 times more likely than H1 (Static≠Moving)
27
*
*
Baseline Static Moving
Sadeghianborojeni et al., Auto-UI 2016, Ann Arbor, MI, USA04.11.2016
Kass, R. E., & Raftery, A. E. (1995). Bayes factors. Journal of the american statistical association, 90(430), 773-795.
27. Time to Collision to Obstacle (secs)
28
F2,38 = 7.70, p < 0.01, ω2 = 0.25
H0 is 4.3 times more likely than H1
(Static≠Moving)
*
*
Baseline Static Moving
Sadeghianborojeni et al., Auto-UI 2016, Ann Arbor, MI, USA04.11.2016
28. NASA RTLX
• Overall Workload (F1.89, 37.95 = 2.16 , p = 0.13)
29
28.57
±16.03
33.21
±12.19
27.14
±16.32
Moving < Static < Baseline
Sadeghianborojeni et al., Auto-UI 2016, Ann Arbor, MI, USA04.11.2016
29. F2,38 = 3.09, p = 0.06, ω2
= 0.09
H0 is 1.3 times less likely than H1 (Static≠Baseline)
H0 is 1.7 times more likely than H1 (Moving≠Baseline)
Number of glances
Baseline Static Moving
30. Glance duration
F2,38 = 2.24, p = 0.12, ω2 = 0.06
Baseline Static Moving
H0 is 5 times less likely than H1 (Static≠Baseline)
H0 is 2 times less likely than H1 (Moving≠Baseline)
32. Findings
• indicating appropriate maneuver
• reduces response times
• increase the safety margin for time to collision
• self-reports indicate less mental demands for moving
cue
• moving cue is not more likely than the baseline
to capture gaze
• Users prefer the moving cue
33Sadeghianborojeni et al., Auto-UI 2016, Ann Arbor, MI, USA04.11.2016
33. Conclusions
34
• Ambient light displays can be effective
in shifting attention to a take-over situation.
• The presentation pattern of the light cues does
not necessarily impair driving performance.
• Presenting contextual information in TORs
can result in more desirable behavior.
Sadeghianborojeni et al., Auto-UI 2016, Ann Arbor, MI, USA04.11.2016
34. Thank you for your attention
shadan.sadeghian@offis.de
lewis@humanmachinesystems.org
Editor's Notes
ford
1. when driving, the driver has to be vigilant to respond for unexpected events that occur rarely, such as the appearance of the pot-hole
2. in addition, he has to concentrate on his primary task, which is driving
3. for driving itself, he might need to switch attention between the traffic ahead and looking at the mirrors for the traffic behind.
4. he might also be talking to a passenger and has to manage how resources are shared between talking and driving
5. naturally, he will want to look at the passenger and extra resources are required to prevent this unhelpful behavior during driving
6. when he notices the sign for the highway exit, he will need resources to prepare the actions for an exiting-highway manoeuvre. this requires resources also.
7. Throughout all of this, he has set himself a goal, namely getting to a specific place, and will need to remind himself of this goal constantly. This requires resources also.
1. when driving, the driver has to be vigilant to respond for unexpected events that occur rarely, such as the appearance of the pot-hole
2. in addition, he has to concentrate on his primary task, which is driving
3. for driving itself, he might need to switch attention between the traffic ahead and looking at the mirrors for the traffic behind.
4. he might also be talking to a passenger and has to manage how resources are shared between talking and driving
5. naturally, he will want to look at the passenger and extra resources are required to prevent this unhelpful behavior during driving
6. when he notices the sign for the highway exit, he will need resources to prepare the actions for an exiting-highway manoeuvre. this requires resources also.
7. Throughout all of this, he has set himself a goal, namely getting to a specific place, and will need to remind himself of this goal constantly. This requires resources also.
first, we show that while having audio cues to prime users
with the urgency of take-over situation, locating the visual
cue in the periphery (namely, a peripheral light display) can
reduce mental workload and assist safe maneuvers
second, our designed light display can convey contextual
information to assist steering at take-over situations
third, using different light patterns for presenting contextual
information can have an effect on driving behavior.
A fixed-based right-hand traffic driving simulator with a field of vision of 150 was used. The simulation was created with
SILAB 1 .
Auditory cues were also played simultaneously from speakers built in the driving
simulator, located behind the driver on both sides.
In this study, an Adafruit NeoPixel Digital RGB LED strip with a resolution of 144 LEDs per
meter was used.
To reduce the intensity of the light display, theLED strip was placed in a matte white acrylic LED profile. The
frame was located on the dashboard of the driving simulator behind the steering wheel, 65 degrees from fixation on the
tablet pc presenting the 1-back task, which was on the drivers’ laps.
To detect the eye gaze of the participants during the experiment, they were asked to wear Dikablis Glasses by Ergoneers
2 . The eye-tracker was calibrated before each trial to ensure constant track of eye-gaze behavior. Two physical markers
on the front panel and two virtual ones on the simulator displays were used.
The calibration procedure took between 30 seconds to one minute for each trial. We used the standard eyetracker
software for calibration, video recording and analysis of participants’ eye-gaze.
In 30-40 second intervals a light and an audio cue was presented as TOR, informing them of a road block (a truck with road construction signs and alerts parked on the road) on either left or the right lane. The TORs were presented at 5 seconds TTC to the road block,
N-back task
Add tablet pic
Measures
Condition (baseline)
Condition (baseline)
Second trial only
Measures
Hypothesis
Post-hoc Tukey HSD tests on both measures revealed
that both static and moving cue conditions were significantly
different from the baseline cue condition but not from each
other.
Post-hoc Tukey HSD tests on both measures revealed
that both static and moving cue conditions were significantly
different from the baseline cue condition but not from each
other.
Post-hoc Tukey HSD tests on both measures revealed
that both static and moving cue conditions were significantly
different from the baseline cue condition but not from each
other.
results
We performed
a JZS Bayesian t-tests in order to understand how the
manipulated cues of static and moving compared to the baseline.
In terms of number of glances, the static cue (BF01=0.21)
was more likely, than the moving cue (BF01=0.49), to be different
from the baseline. In addition, the mean duration of
these glances were more likely to be different for the static
cue (BF01=0.8), than the moving cue (BF01=1.66), to the
baseline.
Using the labels provided by [22], we have ’substantial’ evidence
that static cues attract more glances than the baseline but
only ’anecdotal’ evidence for moving cues. Furthermore, we
have ’anecdotal’ evidence that moving cues result in glances
that have similar duration lengths as our baseline cues, and
’anecdotal’ evidence that static cues result in longer glances.
We performed
a JZS Bayesian t-tests in order to understand how the
manipulated cues of static and moving compared to the baseline.
In terms of number of glances, the static cue (BF01=0.21)
was more likely, than the moving cue (BF01=0.49), to be different
from the baseline. In addition, the mean duration of
these glances were more likely to be different for the static
cue (BF01=0.8), than the moving cue (BF01=1.66), to the
baseline.
Using the labels provided by [22], we have ’substantial’ evidence
that static cues attract more glances than the baseline but
only ’anecdotal’ evidence for moving cues. Furthermore, we
have ’anecdotal’ evidence that moving cues result in glances
that have similar duration lengths as our baseline cues, and
’anecdotal’ evidence that static cues result in longer glances.
first, we show that while having audio cues to prime users
with the urgency of take-over situation, locating the visual
cue in the periphery (namely, a peripheral light display) can
reduce mental workload and assist safe maneuvers
second, our designed light display can convey contextual
information to assist steering at take-over situations
third, using different light patterns for presenting contextual
information can have an effect on driving behavior.