Issues in measuring sensory-motor control performance of human drivers: The case of cognitive load and steering control Johan Engstr

1
Issues in measuring sensory-motor control
performance of human drivers The case of
cognitive load and steering control Johan
Engström, Volvo Technology Corporation European
Workshop on Advanced Predictive Sensory-motor
ControlJoudkrante, Lithuania, 2009-05-21
2
Outline

• Multitasking in the vehicle
• Secondary tasks visual and cognitive
• The primary driving task visual control of
  steering
• Effects of secondary tasks on steering control
• Different effects of visual and cognitive tasks
• The lane keeping improvement effect of
  cognitive load
• Possible explanation in terms of satisficing vs.
  optimising steering control
• Testing predictions implied by this hypothesis

3
Multitasking in the vehicle Driving secondary
tasks
4
Secondary tasks Visual vs. cognitive distraction

• Visual distraction
• Looking off road
• E.g. Visual time sharing when tuning the radio
• Cognitive distraction
• Engaging in demanding cognitive (working memory)
  tasks
• E.g. Mobile phone conversation
• Most real-world tasks involve both components

5
The primary driving task Sensory-motor control
in steering
6
The visual control of steering Optical and
retinal flow
Wann and Wilke (2000)
Straight driving, looking to the left side
Straight driving, looking ahead

• Retinal flow not equal to optical flow

7
Using retinal flow patterns to guide steering
Look where youre going
resulting heading
initial heading
Wann and Wilke (2000)
Underrsteering
Oversteering
Going towards target
8
Gaze angle can be used as a direct cue for
steering through curves (Land, 1998)
Fixate tangent point and adjust steering to
keep gaze angle constant
9
Combining retinal flow patterns and gaze
direction Spring model (Wann and Wilkie, 2000)
Stiffness
Angular acceleration
Damping
Reliance on cues
Main point Foveal vision is essential for
accurate steering!
10
Effects of secondary tasks on steering control
11
Effects of visual distraction on lateral control
Visual time sharing
Gaze angle

• Looking away
• Loosing visual input for steering control
• Heading error builds up
• Looking back
• Large steering wheel correction
• Speed reduction to compensate

Steering wheel angle
Lane position
Increased lane position variance
Speed
Engström and Markkula (2006)
12
What about purely cognitive distraction?

• Large number of simulator and real-world driving
  studies found reduced lane keeping variance
  during cognitive load (Brookhuis et al., 1991
  Östlund et al., 2004 Jamson and Merat, 2005
  Engström et al., 2005 Mattes, Föhl and
  Schindhelm, 2007 Merat and Jamson, 2008).

Engström, Johansson and Östlund (2005)
Does talking on the mobile phone really improve
steering control?
SD lane position
Cognitive task difficulty
13
Other effects of cognitive distraction Gaze
concentration
Victor, Harbluk and Engström (2005)
Normal driving
Cognitive task
14
Other effects of cognitive distraction Increased
steering activity (number of steering reversals
gt 2 deg. per minute)
SW reversals/min
Engström et al. (2005)
15
Summary Effects of cognitive distraction related
to lateral control

• Improved lane keeping (!?)
• Gaze concentration towards the road centre
• Increased number of micro steering corrections
  (lt2 deg)
• How are these effects related? How can they be
  explained?

16
Possible explanation
17
Two key distinctions

• Satisficing vs. Optimising
• Top-down (endogenous) vs. Bottom-up (exogenous)
  attention selection

18
1. Satisficing vs. optimising
Comfort zone
Target value
Satisficing Maintaining performance within
acceptable boundaries.
Optimising Minimising performance error relative
to a target state.
19
Example cost functions of optimising and
satisficing in lane keeping
Cost
Optimising
Satisficing
Lane position
Lane centre
20
Example dynamics of satisficing and optimising
X_dot
Satisficing
Comfort zone
Optimising
X
21
Simulations
Satisficing
Optimising
22
2. Bottom-up and top-down attention selection
Top-down attention bias
Top-down selection
Cognitive task
Other visual task
Vehicle dynamics
Steering
Bottom-up selection
23
Normal driving
Top-down attention bias
Steering easy and automated task,
bottom-up-driven -gt satisficing
Top-down selection
Other visual task
Vehicle dynamics
Steering
Spare top-down attentional resources used for
other visual tasks
Bottom-up selection
visual time sharing

• Lane keeping variance
• Distributed gaze
• Only intermittent steering

24
Cognitive load
Top-down attention bias
Top-down selection
Top-down attention allocated to cognitive task
Cognitive task
Other visual task
No top-down-initiation of other visual tasks
Vehicle dynamics
Steering
Gaze can be fully devoted to steering (attracted
bottom-up)
Bottom-up selection

• Reduced lane keeping
• variance
• Gaze concentration

25
Testable predictions

• General Improved lane keeping should only occur
  if the driver is satisficing in baseline
  condition
• Specific predictions
• Improved lane keeping should not occur if the
  steering task is difficult (so that satisficing
  is not possible)
• Improved lane keeping effect should not occur if
  the driver is motivated to optimise lane keeping
• Support for prediction 1
• Cognitive load has been demonstrated to impair
  performance on tracking tasks (Creem and
  Proffitt, 2001 Strayer and Drews. 2001).
• These tasks could be expected to be more
  difficult and/or less automated than normal
  driving
• Prediction 2 Tested experimentally

26
Instruction to optimise steering (baseline)
Top-down attention bias
Top-down selection
Top-down attention allocated to steering task and
cognitive task
Other visual task
Optimising steering performance
Vehicle dynamics
Steering
Bottom-up selection

• Reduced lane keeping
• variance
• Gaze concentration
• Increased steering
• wheel control input

27
Testing prediction 2 Experimental design

• Simulator study in fixed based simulator (at Saab
  Automobile, Trollhättan)
• Cognitive task Count backwards with 7
• 48 subjects, split in 4 groups
• Incentive for group 1 and 2 Two cinema tickets
  instead of one if meeting some (unspecified) lane
  keeping criterion

Instruction to keep in the middle of the lane Instruction to keep in the middle of the lane
Yes No
Cognitive task Yes Group 1 Group 3
Cognitive task No Group 2 Group 4
28
Prediction

• Lane keeping improvement effect of cognitive load
  should only occur when the driver is not
  motivated to optimise lane keeping satisficing
• Interaction between cognitive load and
  instruction to optimise

29
Preliminary results Lane keeping (HP-filtered SD
Lane Position)
No cognitive task
Effect only for non-instructed subjects
Cognitive task
Due to satisficing in baseline condition
Instructed to optimise lane keeping
No instruction
30
Steering wheel reversal rate
Cognitive task
Same effect in both conditions
Cognitive load less efficient optimising more
steering same lane keeping performance
No cognitive task
Instructed to optimise lane keeping
No instruction
31
Still to be analysed

• Eye movements
• Speed change
• Performance on cognitive task

32
Discussion

• Replicated earlier findings for non-instructed
  drivers
• Reduced lane keeping performance
• Increased steering wheel activity
• Predicted effect of instructions found -gt
  improved lane keeping only for non-instructed
  drivers due to satisficing in baseline
  condition
• Cognitive load seems to induce less efficient
  steering while optimising (more effort in
  steering, same result on lane keeping)
• Cognitive task does not really improve steering
  ability-gt the effect rather reflects
  involuntary improvement from sloppy baseline
  driving

33
General conclusions

• Caution is needed when interpreting driving
  performance measurements do we compare to a
  baseline with satisficing or optimising
  performance?
• In this case, changing instructions and/or
  driving task difficulty may cancel or perhaps
  even reverse the effect of cognitive load
• Implies re-interpretation of many existing
  studies on the effects of cell phone conversation
  on driving performance (e.g. Strayer and Drews,
  2001)