Transcript Green and Slick

Selection of Variables
Paul Green
University of Michigan
Transportation Research Institute
(UMTRI)
Simulator Users Group (SUG) meeting
Session on Behavioral Simulator Fidelity
TRB Annual Meeting
January 13, 2004
What do we want to know?
What is the question we need to answer?
Who are we trying to convince?
Is the road like the real thing?
Is the traffic ...?
Are signs and signals ...?
Is visibility like ...?
Does the vehicle handle ...?
Is the in-vehicle task ...?
Are the drivers…?
Is the environment ...?
Does it sound ...?
The Ultimate Questions
Does the driver behave in the simulator
the same way as on the road?
How large of a difference really matters?
What is the question we need to answer?
What we measure is an answer, so we need
to know what is the question.
*ex1: Does dialing a cell phone while driving
significantly increase the probability of a crash
at signalized intersections?
*ex2: Under what kinds of situations should a person
with a particular stage of Alzheimer’s disease
not be advised to drive?
*ex3: How much hysteresis is acceptable in a
steering system?
Who are we trying to convince?
Scientists and other technical experts ->
technical data presented in scholarly journals
(or proceedings papers or reports)
(publish or perish)
Nonscientists (public at large, legislators, etc)->
Videos / Outtakes
no grainy face shots - reaction is apparent
high quality audio - speech is easily understood
Hands-on drives (if it is just a few people)
(demo or die)
Is the road like the real thing?
AASHTO compliance?
We often study open road driving.
We often measure
Lateral control
is influenced by
(& should be reported)
lane width
curve radius
Longitudinal control posted speed,
natural speed of the road
(shoulders, set backs)
Is the traffic like the real thing?
Amount (ADT or…), not light or heavy
Traffic mix (cars, trucks, SUV/minivans,
motorcycles, pedestrians)
* will matter for following studies
* focus is on cars (but non-cars outnumber cars)
* following large vehicle results in steadier
driving but reduced ability to see ahead
* rarely reported
Vehicle/pedestrian aggressiveness
* we do not know how to measure this
* never reported
Are signs and signals like the real thing?
Almost never
Oversize compensation
MUC TCD compliance (or other)
Signs
Most current projection systems lack the res.
Desired - arc minutes of resolution
usually provide # of chan., FOV, format (XGA)
computable
Signals
not enough luminance or contrast
desired - luminance, contrast
Is visibility like the real thing?
Primary issues
daylight - too dim by orders of magnitude
night driving - no enough contrast for lighting
desired data - mean scene luminance,
mean center stack illumination,
brake light contrast ratio
signs & signals mentioned earlier
Secondary issues
sight distance in fog
Does the vehicle handle like the real thing?
Note: I have ignored motion, an important topic, because I do
not know enough about it to proved informed suggestions.
Case 1: Normal driving (common)
unspecified TLC - railroad tracks problem (measure)
road imperfections, cross winds, tire pressure
excess speed stability
perfectly flat, no head/tail wind
desire: speed decay (rolling resistance)
desire: throttle response (acceleration)
Case 2: Evasive maneuvers (less common)
need input from vehicle dynamics folks on this
(lots of HFEs are psychologists)
Is the in-vehicle task like the real thing?
Often
Touch screen vs. real device
Ease of instrumentation
No suggested measures of fidelity
(task measures are time and errors)
Are the drivers (subjects) like of real drivers?
I test students because…
They are easier to get
They cost less
They understand the instructions
There are fewer IRB concerns
Their performance is less variable
If your question concerns drivers of all ages,
especially if it relates to the use of technology,
include older subjects (e.g., over 65) in the sample.
Does it sound like the real thing?
Prior UMTRI work showed sound has little effect
on basic driving performance.
Modern motor vehicles are quiet.
Only usually matters (masking)
auditory warning
speech I/O
No measures suggested
Is the environment like the real thing?
Probably more of an issue for military
and off-road driving (trees, buildings, etc.)
No measures suggested
Does the driver behave in the simulator
the same way as on the road?
Sometimes yes, sometimes no
No consequence of leaving lane
No consequence of poor speed control
No consequence of poor situation awareness
Suggested Driving Performance Measures
It really depends on the question…
But here is what is sometimes reported.
(Comment on each - ease of measurement,
crash/behavioral implications)
Mean speed
Standard deviation of speed
Speed drops during task
Less common
Throttle variance
Throttle power spectrum (holds)
Lane departures (the definition problem aside)
What constitutes a lane departure? (12+ combos)
1. Which part of the vehicle matters?
outer edge of side mirrors
outside curve of body
outside edge of tire (contact or belly)
tire centerline
2. Where on the line?
touching the nearest edge line
centerline of the edge
past the farthest part of the edge line
Lane departures
Standard deviation of lane position
Less common
TLC (mean or some other %)
%tile of lane position
headway distance or time
Headway variance
Less common
Coherence measures
TTC
Standard deviation of steering wheel angle
Less common
Steering entropy
Steering power spectral distribution & related measures
How large of a difference really matters?
Practical vs. statistical significance
Other impairments - “It’s like being drunk.”
Do not forget to report means
Do not forget about significant figures
(What does a standard deviation of lane
position of .213 m mean?)
The problem is that we do really know which
measures are “best,” in part, because we do not
have a theory that describes how people drive.
We need a science of driving.
Summary: What Must We Report?
(Green’s List)
If we ask for too much, people will not do it.
Subjects
Age
Vehicle
TLC uncontrolled
Width (for lane departure)
Speed variability at fixed throttle angle (speed specific)
Speed decay constant
Acceleration at some partial throttle
Road
Lane Width
Shoulder width/distance to barrier
AASHTO/MUTCD/etc. compliance
Traffic
ADT
Mix
Aggressiveness
Lighting
Scene resolution (min of arc)
Mean scene luminance
Traffic light/Brake light contrast ratio (& luminance)
Mean center stack illumination (if relevant)
Any Questions?