Collision Warning Design To Mitigate Driver Distraction (CHI 2004)

1
Collision Warning DesignTo Mitigate Driver
Distraction (CHI 2004)

• Andrew Muller
• Eugene Khokhlov

2
University of Iowa

• Elizabeth
• Hayes
• Daimler Chrysler
• (Chewbacca)

• Joshua
• D. Hoffman
• Grad Student

• John D. Lee
• Ph.D.

3
To The Point

• The Problem Too many distractions while driving
  a car
• The Need Collision warning system

4
Background Information

• In-vehicle Information Systems (IVIS) are now
  feasible because
• Technology Advances
• Societal Trends
• IVIS Functionality
• Response Types
• Critical Factors for IVIS

5
Alert Strategies

• Warning Strategies
• Graded
• Single Staged
• Sensor Modality Presentation
• Haptic (touch)
• Auditory

6
Experiment Goals

• Experiment 1
• Examine how driver response depends on graded and
  single stage warnings
• Examine how driver response depends on modality
  (haptic vs. auditory) of the warning
• Experiment 2
• Examine how these warning strategies and
  modalities affect driver preference

7
Experiment 1 Method

• A mixed between/within-subject experimental
  design
• 3, 15-minute driving scenarios
• 21 braking events (7x3)21
• 3 levels of severity
• Speech-based email system to distract the driver

8
Participants

• 40 individuals
• 20 female, 20 male
• Ages of 25 and 55 (licensed)
• Unaware of the nature of the research
• Paid 20 each

9
Apparatus

• Fixed-based, medium-fidelity driving simulator
• 1992 Mercury Sable
• 50-degree visual field of view
• 640x480 screen
• Visual collision warning icon
• Needed elements for auditory and haptic alerts

10
DriveSafety (Hyperion)
11
Experimental Design and Independent Variables

• Mixed between-within subject design
• Between subject variables
• Warning modality
• Warning strategy
• Within subject variables
• Severity of lead vehicle breaking
• If response was require

12
Dependent Variables

• Safety benefit
• Number of collisions
• Adjusted minimum time to collision (AMTTC)
• Driver response process (response followed by
  assessment or assessment followed by response)

13
Procedure

• Operation instruction
• Introductory drive (5 min)
• 3 main drives (15 min/each)
• 7 braking events per drive
• _at_ 55mph
• 1/7 was severe, always at end
• Complete auditory email task

14
Results

• 741 data points total
• Repeated-measures ANOVA was used to analyze the
  data using two-tailed hypothesis tests

15
Results Severity of braking events and driver
response

• Drivers responded to braking events in a
  systematic and realistic manner
• AMTTC reflected braking severity
• Severity of lead vehicle braking affected
  drivers braking response
• Severity of braking affected mean deceleration

16
Results Interface characteristics and safety
benefit (collisions)

• 40 potential collisions
• 10 collisions occurred
• 7 in single-stage and 3 in graded
• X2(1) 2.13, p0.144
• 5 in auditory and 5 in haptic

17
Results Interface characteristics and safety
benefit (AMTTC)

• Slight benefit for graded compared to
  single-stage
• F(1,36)8.74, p0.0055
• Graded substantially better in severe braking
  events

18
AMTTC
19
Response to nuisance alarm braking events
20
Experiment 2 Method

• A within-subject experimental design
• 4, 10-minute scenarios
• 24 braking events
• 3 levels of severity
• 2/3 of events required no driver response

21
Participants

• 20 individuals
• 11 females, 9 males (licensed)
• Between the ages of 25 and 55
• Unaware of the nature of the research
• Paid 20 each

22
Apparatus Independent variables and
experimental design

• Same as in experiment 1

23
Dependent variables

• Driver attitudes were measured with a series of
  subjective rating scales after each drive
• After completion of all trials, they
  comparatively ranked the systems

24
Procedure

• Operation instruction
• Introductory drive (5 min)
• 4 main drives (10 min/each)
• 6 braking events per drive
• Each scenario had an equal number of event
  severity

25
Results

• Rank the warning modalities in order from 1 to 4
  based on preference
• Violation of assumption of a repeated measures
  ANOVA
• Applied Friedmans non-parametric analysis
• Only when Friedmans showed a significant
  difference between conditions was a post-hoc
  multiple comparison performed using Fishers
  least significant difference method

26
(No Transcript)
27
Conclusions

• Graded warning provided a greater safety margin
• Graded warning induced fewer inappropriate
  responses to the nuisance alarms
• Graded warning was more trusted
• Warning modality had little effect on performance
  in severe braking events
• Haptic warnings were preferred on several
  dimensions to auditory

28
Questions

• In table 2, graded haptic beats single-stage
  haptic in everything except overall preference,
  what can account for this?
• Does the data on table 2 match what you would
  have expected?
• Graded is preferred for a one hour experiment,
  how about 5-10 years on daily basis?

29
Questions

• Haptic is preferred over auditory in this study,
  is this a property of auditory or a property of
  the time span of the test, or some other factor?
• Why express haptic through a seat and not a gas
  pedal as in previous studies?