IVBSS Program Overview
Jim Sayer
Program Overview and Purpose
• A 5 year effort to develop and field test integrated
safety systems for passenger vehicles and
heavy trucks
• Safety systems provided warnings for rear-end,
lane change and road departure crashes in
addition to arbitration of multiple crash threats
• Included a one-year field test on public roads, and
analyses
y
determined system
y
p
performance,, safety
y
benefits, and user acceptance
October 20, 2010
2
Development
p
Team
October 20, 2010
3
Federal Partners
U.S. Department of Transportation
Intelligent Transportation Systems
Joint Program Office
October 20, 2010
4
Timeline - Vehicle Deployment
Phase II – 30 months
Phase I – 30 months
Nov 2005
Nov 2006
Engineering
Development
Vehicles
October 20, 2010
June 2008
Prototype
Vehicles
Nov 2008
Pilot
Vehicles
Feb 2009
Extended
Pilot
FOT
Nov 2010
FOT
Data
Collection
5
IVBSS Passenger Vehicle
Radar
Lane-change/Merge
((LCM))
Vision
Lateral Drift
Warning (LDW)
Curve
speed
Warning
(CSW)
October 20, 2010
Forward Crash
W i (FCW)
Warning
6
Light Vehicle Integration
October 20, 2010
7
Heavy Truck Sensor Coverage
Radar*
Vi i *
Vision*
40˚
Lane-change/Merge
(LCM)
100˚
100˚
Forward Crash
Warning (FCW)
11˚
38˚
˚
11˚
40˚
100˚
100˚
Laterall Drift
f
Warning (LDW)
* Sensor field-of-view
field of view depth not to scale
October 20, 2010
8
Heavy Truck Integration
October 20, 2010
9
Light Vehicle Participants
• 108 d
drivers
i
– younger (20-30 years)
– middle-aged (40-50 years)
– older (60-70 years)
• 16 IVBSS vehicles
• 6 weeks of driving per participant
First 12 days – Warning inhibited (Baseline)
y – IVBSS warnings
g enabled ((Treatment))
Next 27 days
October 20, 2010
10
Light Vehicle FOT
• April
April, 2009 to April
April, 2010
• Data Set
– 213,000 miles
– 23,000
23 000 trips
– 6,200 hours
– 600 d
data
t channels
h
l
– 5 video channels
October 20, 2010
11
Light Vehicle Video Example
October 20, 2010
12
Heavy Truck Participants
• 20 drivers initially recruited
– 8 Pick-up and Delivery (daytime, short trips)
– 10 Li
Line-Haul
H l ((nighttime,
i htti
llong ttrips)
i )
– Average age:
• P&D – 48 years
• LH – 50 years
– Average commercial driving experience:
• P&D – 18 years
• LH – 25 years
October 20, 2010
13
Heavy Truck FOT
• February,
February 2009 to December
December, 2009
• Data Set
– 602,000 miles
– 23,000
23 000 trips
– 14,000 hours
– 500 d
data
t channels
h
l
– 5 video channels
October 20, 2010
14
Heavy Truck Video Example
October 20, 2010
15
IVBSS Reports and
D
Documentation
t ti
• 28 technical reports
– Over 3000 pages of documentation
– Functional
F
ti
lR
Requirements
i
t and
dS
System
t
P
Perf.
f
– Verification Testing
– Annual reports
• Electronic posting
– www.umtri.umich/ivbss.php
– www.nhtsa.gov/research/crash+avoidance
October 20, 2010
16
Structure of Final Reports
Posted
Heavy Truck Key
Findings Report
Light Vehicle Key
Findings Report
To be posted
before 10/30
Heavy Truck
Methodology &
Results
Light Vehicle
Methodology &
Results
IVBSS Final Report
October 20, 2010
In Final Review
Draft 10/27
Draft Submitted
Complete 11/12
17
Wrap Up
Wrap-Up
• Presentations from today will be posted
• UMTRI willing to give presentations on the
program
• Data will be made available for research
• Research vehicles located outside
• Representatives from all partners present
October 20, 2010
18
Heavy Truck Key Findings
Scott Bogard
Contributors: J. Sayer, D. Funkhouser,
A Blankespoor,
A.
Bl k
S.
S B
Bao
M. L. Buonarosa, D. LeBlanc
IVBSS Heavy Trucks
• 10 vehicles with three
prototype crash warning
systems
• 8 radars, 5 video streams,
GPS, >500 other signals at
10
0 to 50 Hz
• Eaton & Takata with
support from Navistar
• Con-way Freight’s Detroit
terminal was the fleet
October 20, 2010
20
Heavy Truck Driver Vehicle
I t f
Interface
October 20, 2010
21
Heavy Truck Participants
• 20 drivers initially recruited
– 8 Pick-up and Delivery (daytime, short trips)
– 10 Li
Line-Haul
H l ((nighttime,
i htti
llong ttrips)
i )
• 10 months of driving per participant
– 2 months – IVBSS warnings inhibited (Baseline)
– 8 months – IVBSS warnings enabled (Treatment)
• Test period: Feb., 2009 to December, 2009
October 20, 2010
22
Geographical range of driving
b P&D d
by
drivers
i
October 20, 2010
23
Geographical range of driving
b liline-haul
by
h ld
drivers
i
October 20, 2010
24
All IVBSS Warnings
• 85
85,933
933 warnings (treatment mode
mode, shown
below)
FCW (15397)
18%
LCM (9317)
58%
11%
14%
LDW-Imminent
(11720)
LDW-Cautionary
(49499)
October 20, 2010
25
Warning Events & Driver Experience
• Lateral drifts toward unoccupied space:
– Accounts for 72%
– 98% of these warnings
g are considered valid
• Lane change merge and Lateral drift warnings
toward an occupied space:
– 33% are considered valid
• Forward collision warning (FCW):
– 88% of warnings to targets that were seen moving were
valid
– 52% of FCWs were triggered by fixed objects
October 20, 2010
26
HT Comprehensive System Results:
Di
Driver
B
Behavior
h i
• There was no effect of the integrated system
on frequency of secondary tasks.
– 1,980
1 980 sampled
l d video
id clips
li were coded
d d ffor
secondary tasks
– Slight reduction from 43
43.5%
5% to 42
42.1%
1%
– P&D drivers less likely to engage in secondary
tasks
October 20, 2010
27
HT Comprehensive System Results:
Di
Driver
B
Behavior
h i
– In multiple threat scenarios
scenarios, the initial warning
was generally enough to get the attention of
drivers
• 83 cases observed were during treatment mode
October 20, 2010
28
HT Comprehensive System Results:
Di
Driver
A
Acceptance
t
• Drivers believe that the integrated system
will increase their driving safety
October 20, 2010
29
HT Comprehensive System Results:
Di
Driver
A
Acceptance
t
• Drivers rated the integrated system fairly
well in terms of usefulness and satisfaction
Satisfa
action
0 70
0.70
LDW
0.60
0.50
0.40
0.30
IVBSS
BSD
0.20
FCW
0.10
0.00
0.00
LCM
0.50
1.00
1.50
Usefulness
October 20, 2010
30
HT Comprehensive System Results:
Di
Driver
A
Acceptance
t
• 15 of the 18 drivers prefer a truck with the
integrated system to a conventional truck
October 20, 2010
31
HT Lateral Control and Warnings:
Di
Driver
B
Behavior
h i
• The integrated system did not have an effect
on reducing lane departures
– But
B t departures
d
t
decreased
d
d ffor 13 off 18 d
drivers
i
• The duration and distance of lane incursions
was nott affected
ff t d by
b the
th integrated
i t
t d system
t
• There was no effect of the integrated system
on turn-signal use during lane changes
October 20, 2010
32
HT Lateral Control and Warnings:
Di
Driver
B
Behavior
h i
• The integrated system had a statistically
significant, but practically very small, effect
on lateral offset
– Drivers were 1.7 cm closer to the center of the
lane
SV
October 20, 2010
Lateral Offset
33
HT Lateral Control and Warnings:
Driver Behavior
• The duration and distance of lane incursions
was not affected by the integrated system
• There
Th
was no effect
ff t on the
th frequency
f
off
lane changes
October 20, 2010
34
LDW/LCM Example Videos
October 20, 2010
35
HT Longitudinal Control and Warnings:
Di
Driver
B
Behavior
h i
• No effect of the integrated system on
forward conflict levels
• Drivers
Di
maintained
i t i d marginally
i ll llonger
average time headways (0.05s)
• No effect on of hard-braking events (less
than 0.2g [1.96 m/s2])
• Drivers responded more quickly to closingconflict events in the treatment condition
October 20, 2010
36
HT Longitudinal Control and Warnings:
Di
Driver
B
Behavior
h i
• No significant effect on gap size when
performing lane changes.
October 20, 2010
37
FCW Example Videos
October 20, 2010
38
HT Longitudinal Control and Warnings:
Di
Driver
A
Acceptance
t
• Line
Line-haul
haul and P&D drivers specifically
mentioned valid FCW warnings and the
headway time margin display to be helpful
headway-time
• Driver acceptance, while favorable, would
almost
l
t certainly
t i l h
have b
been hi
higher
h h
had
d
invalid warnings due to fixed roadside
objects
bj t and
d overhead
h d road
d structures
t t
October 20, 2010
39
Summary – HT FOT
• Drivers generally accepting of the integrated
system
• Drivers
Di
reported
t d th
thatt th
the LDW system
t
was
most beneficial
• Secondary task activity unchanged
• Multiple-threat
p
warning
g sequences
q
are rare
October 20, 2010
40
Independent Evaluation
Res lts – Heavy
Results
Hea Tr
Truck
ck
IVBSS Public Meeting
October 20, 2010
Presentation Outline
•
•
•
•
•
42
Evaluation status
Goals and objectives
y approach
Analysis
Study limitations
Heavyy truck results
– Safety impact
– Driver acceptance
– System performance
Evaluation Status
• Draft final report submitted August 9, 2010.
• Comments
C
t received
i d early
l O
October,
t b 2010
2010.
• Final report due October 29, 2010.
43
Independent Evaluation Goals
44
Safety Impact Framework
Overall Driving
Driving Performance
Driver Inattention
Frequency of Occurrence
Driving Conflicts
Near
Crashes
Driver Response
Frequency of Occurrence
Driver Inattention
45
Projection
of
Potential
Safety
Benefits
Overall Driving Measures
• Driving Performance
–
–
–
–
Travel speed
Time headway
Lane change maneuvers
Lane keeping
• Driver
Di
IInattention
tt ti
– Secondary tasks
– Eyes-off-forward-scene
Eyes off forward scene
46
Driving Conflict Measures
• Exposure
– Rear-end conflicts
• Lead vehicle decelerating/moving at slower speed
• Lead vehicle stopped
– Lateral conflicts
• Lane change
• Road departure
• Response
– Rear-end conflicts: time-to-collision/deceleration level
– Lateral conflicts: lateral acceleration/lane bust distance
47
Near Crashes – Thresholds
• Rear-end driving conflicts:
– TTCmin < 3 sec
– Ax
A min > 0.2g
02
– Brake pedal > 0.5 sec
OR
• Lane
Lane-change
change conflicts:
– Aymax > 0.1g
– 0 m < dLBmax < 0.9 m
• Road-departure conflicts:
– Aymax > 0.1g
– 0.3
0 3 m < dLBmax < 0.9
09m
– 1 sec < tLB < 5 sec
48
OR
– Axmin > 0.4g
– Brake pedal > 0
0.5
5 sec
TTCmin: minimum time-to-collision
A min: minimum
Ax
minim m deceleration le
level
el
Aymax: maximum lateral acceleration
dLBmax: maximum lane bust distance
tLB: lane bust duration
– Aymax > 0.3g
– 1 sec < tLB < 5 sec
Video Analysis
All Alerts
Al t
Video Available
Distraction
Steering Response
Host Vehicle Position
Crash Imminent
Eyes off Forward Scene
Host Vehicle Maneuver
Location
FCW Alerts
T
Target
Type
T
Target Vehicle Body Type
Lead Vehicle Maneuver
Lead Vehicle Position
In path of host Vehicle
LDW-I /LCM Alerts
Target
g Type
yp
Target Location
Moving Target Vehcicle Relative Speed
LDW-I/LDW-C
Lane Excursion Scenario
Lead Vehicle Maneuver Times
Lane Marker
R dC
Road
Condition
diti
Opposing Traffic
FCW: Forward Crash Warning
Time of Collision
LCM: Lane Change/Merge
p
Warning/Imminent-Cautionary
g
y
LDW/I-C: Lane Departure
49
Study Limitations
• Professional drivers participating in the study
were generally very safe, even during the
baseline period
• Very low exposure to some near crash types
i b
in
both
th ttestt periods
i d prevented
t d safety
f t
projections for 4 pre-crash scenarios
• Small
S ll sample
l off ttestt subjects
bj t
50
Pre‐Craash Scenario
Target Pre-Crash Scenarios
Rear‐end/lead vehicle moving
5.7%
Drifting/same lane
5.7%
Opposite direction/no maneuver
6.5%
Rear‐end/lead vehicle decelerating
8.9%
Road edge departure/no maneuver
Road edge departure/no maneuver
14 6%
14.6%
Rear‐end/lead vehicle stopped
15.4%
Turning/same direction
19.5%
Changing lanes/same direction
23.6%
0%
5%
10%
15%
20%
Portion of Involved Heavy Trucks
Portion of Involved Heavy 123,000 heavy trucks were involved in target
pre-crash scenarios annually
(based on average 2004-2008 GES statistics)
51
25%
Overall Driving – Turn Signal Use
• Increased turn signal usage for line-haul drivers
(speeds over 45 mph)
– Baseline: 78% of lane changes signaled
– Treatment: 82% of lane changes signaled
– 5% increase
52
Overall Driving – Lane Busts
• Decrease in rate of lane busts to both sides
– 9% reduction at slower speeds (35 to 55 mph)
• Baseline: 8.7 per 10 miles, Treatment: 7.9 per 10 miles
– 15% reduction at higher speeds (above 55
mph)
• Baseline: 3.5 per 10 miles, Treatment: 3.0 per 10 miles
– Most improvement to the right side
53
Overall Driving – Secondary Tasks
• 8 line-haul drivers showed an increase in secondaryy tasks
during the treatment period.
• No negative safety impact is associated with this increase.
Change iin % of Episodes
s with Secondary Tasks
25%
LH: Line-Haul
21%
20%
17%
15%
LH
9%
10%
P&D
5%
2%
4% 4% 4%
3%
10%
6% 6%
0%
5
-5%
1
10
7
2
4
6
8
22
-2% -1%
-3% -3%
29
-1%
-5%
5%
-10%
-15%
54
P&D: Pickup & Delivery
-12%
28
27
25
26
21
24
30
23
Secondary Tasks – Change from Baseline to Treatment
Pickup & Delivery
Driver
55
Line-Haul
Secondary Task
Driver
Secondary Task
21 Eating
E i
12%
1
Looking outside car
6%
2
Smoking
8%
22
Looking outside car
13%
4
Grooming
6%
23
Eyes Closed >1s
16%
5
Reaching for object
3%
24
Eyes Closed >1s
8%
6
Talking on phone
5%
25
Reaching for object
6%
7
Looking outside car
9%
26
Smokingg cigarettes
g
9%
8
Grooming
9%
27
Groming
7%
10
Smoking
10%
28
Talking on Phone
9%
29
Looking outside car
10%
30
Talking on bluetooth
19%
Overall Driving – Eyes-Off-Forward Scene
No trend in eyes-off-forward scene behavior with system enabled
Change in % of Episodes with Eyes off
Fo
orward Scene (T-B)
20%
15%
7%
% 3%
2% 3%
5%
4
-25%
8
-3%
-10%
-15%
2% 3%
0%
0%
-5%
17%
10%
10%
-20%
20%
56
15%
-11%
6
1
2
5
10
7
22
24
26
3%
-3%
-8%
P&D
P&D: Pickup & Delivery
-17%
-18%
18%
25
-6%
21
29
27
23
30
-4%
LH
LH: Line-Haul
28
Overall Driving – Eyes Closed
• 502 instances of “Eyes closed for > 1 second” observed
• 3 line-haul drivers accounted for 90% of instances
• D
Driver
i
21
21: 7% off b
baseline
li alerts,
l t 12% off ttreatment
t
t alerts
l t
• Driver 23: 19% of baseline alerts, 35% of treatment alerts
• Driver 24: 5% of baseline alerts, 13% of treatment alerts
Po
orportion of A
Alerts 100%
Numbers indicate total
number of observations
80%
60%
P&D: Pickup & Delivery
LH: Line-Haul
40%
281
20%
77
1
1
2
1
1
2
4
5
98
17
1
5
18
25
26
27
0%
P&D
57
6
7
8
10
21
Driver
22
23
24
LH
28
29
30
Driving Conflict Results
• No statistically significant reduction in driving
conflicts
– Route type
– All treatment
– Last ¼ of treatment mileage
– Last 2 hours of shift
58
Near Crash Experience
•
•
•
12 drivers experienced a decrease in near crash rates
Reduction in rates of road departure to left: 2.0 to 1.1 per 1,000 miles
(statistically significant)
Reduction in overall near crash rates: 8.7 to 8.1 per 1,000 miles (not
statistically significant)
Chaange in Near Crash EEncounter per 1000 Mile
es
80%
67%
60%
44%
40%
32%
26% 26%
20%
6%
3%
0.4%
19%
12% 12%
13%
0%
1
2
4
5
6
‐20%
‐40%
7
8 10 21 22 23 24 25 26 27 28 29 30
‐7%
‐24%
P&D
‐60%
‐80%
P&D: Pickup & Delivery
59
← Decrease
‐22%
‐33%
‐44%
44%
LH
← Increase
‐67%
LH: Line-Haul
Near Crash Experience Reported by Subjects
• 8 drivers reported that the system prevented them
from getting into a crash or near crash
– Pi
Pickup
k &D
Delivery
li
d
drivers:
i
• Lead vehicle cut-in (2 drivers)
• Slower moving lead vehicle
• Vehicle in blind spot when attempting lane change
– Line
Line-Haul
Haul drivers:
• Slower moving lead vehicle
• When tired and drifting off the road
• Vehicle in blind spot when attempting lane change
• Encountered lead vehicle with no brake lights
60
Driver Acceptance – Key Results
• 15 drivers would prefer to drive a truck with the
integrated system
system.
• 13 drivers felt that driving with the integrated
system would increase their driving safety.
safety
• 15 drivers reported that the system made them
more aware of their surroundings
surroundings.
61
Driver Acceptance
What did you like most about the integrated system?
Pickup
Pick
p & Deli
Delivery
er
Increased awareness
Line-Haul
Line
Ha l
Blind spot monitors
(3 drivers)
(4 drivers)
Lane departure warnings
(3 drivers)
62
Driver Acceptance
What did you like least about the integrated system?
63
Pickup
Pick
p & Deli
Delivery
er
False FCW warnings
Line-Haul
Line
Ha l
False side hazard warnings
(5 drivers)
(4 drivers)
Driver Acceptance
In which situations did you find the integrated system
to be most helpful?
Pickup & Delivery
Approaching slower traffic
Line-Haul
Drifting
(2 drivers)
(4 drivers)
Carss in blind
Ca
b d spot
(2 drivers)
64
Driver Acceptance
Reported changes in driving behavior
Increased
alertness/
attention
(3 P&D, 1 LH)
None
(3 P&D, 8 LH)
LH: Line-Haul
P&D: Pickup & Delivery
65
Increased
headway
(2 P&D)
Improved lane
positioning
(1 LH)
Driver Acceptance
Did you rely on the integrated system?
Yes
LH
4
P&D
No
6
1
7
LH: Line-Haul
P&D: Pickup & Delivery
0
66
2
4
6
8
Number of Responses
10
Driver Acceptance
Did you rely on the integrated system?
• “Yo
“You find yourself
o rself rela
relaxing
ing a bit
bit.””
• “To a small degree, but habit and experience led
me to rely on myself most of the time
time.”
• “For lane changes in bad weather and sunlight, I
didn’tt check my mirrors until the [BSM] went off
didn
off.”
• “There were times when I was very tired, the drift
warnings were helpful
helpful.”
• “I found it helped keep me alert in heavy traffic.”
67
System Performance – Driver Vehicle Interface
Positive
Neutral
Negative
Center display
p y
Center Display Useful
12
Blind
d spo
spot
lightsBSM Lights Useful
3
11
Blind spot
BSM light
Displayed
Convienently
location
2
9
0
2
4
5
4
5
6
8
10
12
Number of Responses
68
3
14
16
18
System Performance – Auditory Warnings
Positive
Neutral
I could easily
distinguish LH
among the
auditory warnings
Negative
17
The auditory
P&D
warnings got my
attention
1
16
0
2
4
6
8
1
10
Number of Responses
69
12
14
16
1
18
System Performance – Controls
9 drivers adjusted the
volume more than 5 times
Positive
Volume
Volume Control
Control
9
Mute
Mute Button
Button
2
Only 2 drivers used the mute
button more than twice
Negative
g
3
6
5
6
0
70
Neutral
4
6
8
7
10
Number of Responses
12
14
16
18
System Performance – Forward Crash Warning (FCW)
• Forward radar characterization
•
•
93% of FCW-moving alerts issued for in-path target
3% of FCW-stopped
FCW stopped alerts issued for in-path
in path target
Prop
portion of Out‐O
Of‐Path Target Aleerts
FCW‐M
100%
93%
FCW‐S
98%
97%
90%
80%
70%
60%
50%
40%
30%
20%
10%
5%
10%
7%
0%
P&D
LH
(Based on video review of 2,368 FCW alerts)
71
All
System Performance – Breakdown of FCW
Al t Issued
Alerts
I
d for
f Out-of-Path
O t f P th Stationary
St ti
Objects
Obj t
Prop
portion of Out‐Of‐Path Targets
70%
64%
P&D
60%
50%
44%
40%
30%
22%
20%
10%
15%
13%
9%
11%
8%
5%
0%
0%
No Visible Target
Roadside sign
Bridge
Guardrail
(Based on video analysis of 875 FCW-stopped alerts)
72
Pole
LH
System Performance – FCW-M Alert Experience
•
•
73
12% reduction in FCW-moving alert rate from baseline to treatment
• 24% reduction for line-haul drivers (7.0 to 5.2 per 100 miles)
Indicates improved headway keeping for all drivers, especially line-haul drivers.
System Performance – Side Imminent Alerts
•
•
•
58% of LCM alerts had adjacent target
56% of LDW-I alerts had adjacent target
Side detection more accurate in p
pickup
p & delivery
y driving
g environment
Proportion of N on‐Adjacent Taarget P
Alerts
A
LCM
LDW‐I
60%
52% 51%
50%
42%
37%
40%
30%
(Based on video review
Of 6,861
O
6,86 a
alerts)
e ts)
25%
20%
10%
0%
P&D
LCM: Lane change/merge warning
74
44%
LH
All
LDW-I: Imminent lane departure warning
System Performance – Breakdown of NonAdj
Adjacent
t Target
T
t Types
T
in
i Side
Sid Imminent
I
i
t Alerts
Al t
Proporrtion of Non‐Adjjacent Targets
70%
60%
P&D
59%
53%
50%
40%
30%
23%
20%
16%
20%
8%
10%
9%
4%
0%
No Visible Target
Moving Vehicle
Stationary Vehicle
Roadside sign
(Based on video analysis of 2,743 alerts)
75
LH
System Performance – Validity of Cautionary
L
Lane
Departure
D
t
W
Warning
i (LDW-C)
(LDW C) Al
Alerts
t
•
•
90% of LDW-C alerts issued for valid lane departure
Higher accuracy for line-haul drivers (more freeway miles)
Pro
oportions of LDW
W‐C Alerts by LLane Excu
ursion
Unintentional
100%
90%
13%
7%
None
10%
13%
80%
70%
Intentional
24%
39%
60%
50%
40%
30%
20%
79%
67%
49%
10%
0%
P&D
LH
All
(Based on video analysis of 3,671 LDW-C alerts)
76
System Performance – LDW-C Alert Experience
•
•
77
20% reduction in LDW-C alert rate from baseline to treatment
• 14% reduction for pickup & delivery drivers (6.9 to 6.0 per 100 miles)
• 25% reduction for line-haul drivers (7.0 to 5.2 per 100 miles)
Indicates improved lane keeping for all drivers with the integrated system
enabled
System Performance – Steering Intensity
• Increase in corrective steering intensity in response to
side alerts when audible warning was issued
0.90
0.80
0.70
Maxx Ay (m/s^2)
0.60
0.50
0.40
0.30
0.76
0.69
0.67
0.61
0.59
0.84
0.79
0.75
0.20
0.10
0.00
Baseline
Treatment
LCM
78
Baseline
Treatment
LDW‐I
Baseline
Treatment
LDW‐C
Baseline
Treatment
All
System Performance – LDW Availability
• Lane tracking availability met target performance
specifications for all speed bins
Proportio
on of Miles Travveled
25 ≤ V < 35
100%
90%
80%
70%
60%
50%
40%
30%
20%
10%
0%
35 ≤ V < 55
89%
87%
75%
46%
Left Side
Left Side
V ≥ 55
69%
40%
Right Side
Right Side
LDW Availability
79
85%
62%
34%
Both Sides
Both Sides
Target
Values
Projection of Potential Safety Benefits
Function
80
Pre Crash Scenario
Annual Target
C h
Crashes
Max Estimated
Annual Crash
Reduction
Max Estimated
Eff ti
Effectiveness
5,000
27%
FCW-M
Rear end/Lead vehicle decelerating
Rear end/Lead vehicle moving
18,000
FCW-S
Rear end/Lead vehicle stopped
19,000
LCM
Changing lanes/same direction
Turning/same direction
53,000
LDW I
LDW-I
D ifti /
Drifting/same
lane
l
7 000
7,000
LDW-C Left
Opposite direction/No maneuver
Road edge departure/No maneuver
11,000
3,000
29%
LDW C Right
LDW-C
Ri ht
R d edge
Road
d departure/No
d
t /N maneuver
15 000
15,000
5 000
5,000
36%
Integrated System
All
123,000
13,000
11%
Insufficient field data to estimate
Recommendations
•
81
System improvements for more accurate detection of:
• Stationary forward targets
• Adjacent targets
Questions
82
Light Vehicle Key
Findings
g
Dave LeBlanc
Contributors: J. Sayer, M. L.
Buonarosa D. Funkhouser, S. Bao, A.
Blankespoor S
Blankespoor,
S. Bogard
IVBSS Light Vehicles
• 16 vehicles with four
prototype crash systems
• 7 radars
radars, 5 video
cameras, GPS, >600
channels at 10 to 50 Hz
• Visteon & Takata with
support from Honda
Radars
October 20, 2010
behind fascias
Lane-change/Merge
Radar
(LCM)
Radar
sLane-tracking
Vision
Curve
speed
Warning
(CSW)
Lateral Drift
Warning (LDW)
Forward Crash
W
Warning
i (FCW)
84
Light Vehicle Driver Vehicle
I t f
Interface
Forward crash warning
Audible – forward tone
Audible – forward tone
Haptic – brake pulse
Text to confirm: “Sharp Curve”
Text to confirm: “Forward Hazard”
Curve speed warning
October 20, 2010
85
Light Vehicle Experiment
• 108 drivers
–
–
–
–
Younger (20-30 years)
Middle-aged
Middle
aged (40-50
(40 50 years)
Older (60-70 years)
Same number of males and females in each age group
• 6 weeks of driving per participant
– First 12 days – IVBSS warnings inhibited (Baseline)
– Next 27 days – IVBSS warnings enabled (Treatment)
• Test period: April 6, 2009 to April 20, 2010
October 20, 2010
86
Travel in Light Vehicle FOT
213,309
miles
CSW map coverage
22,657
Percenttrips
of
time in motion
6164
hours
ADAS map:
p 76.9%
2
terabytes
non ADAS map: 15.5%
No map: 7.6%
October 20, 2010
87
IVBSS:
7C
Crashes
h (N
(No IInjuries)
j i )
• One driver struck the rear
rear-end
end of another vehicle
• One driver ran off the road (~30 mph)
• One
O driver
di
hit a construction
t ti b
barrell d
during
i
a llastt
moment move to exit a freeway
• Two drivers backed into stationary objects (a
parked vehicle, a fence)
• Two drivers struck running deer at night
• Impact, not crash: curb strike (damages fascia)
October 20, 2010
88
Travel in LV FOT
• 213,309
213 309 miles (32% in baseline)
• Males accounted for 61% of travel distance
Middle-aged had most
travel distance…
Half the travel distance was on
limited access & ramps
Younger
29%
32%
Middle age
Limited access
Ramps
9%
45%
Surface
Older 42%
Unknown
39%
4%
October 20, 2010
89
All IVBSS warnings
• 10
10,617
617 warnings (baseline)
• 12,202 warnings (treatment mode, shown
b
l )
below).
5% 5%
FCW (579)
7%
13%
70%
LCM (915)
LDW-Imminent
(1593)
LDW-Cautionary
(8505)
CSW (610)
October 20, 2010
90
Warning Events & Driver
E
Experience
i
• Lateral drifts toward unoccupied space:
– Accounts for 77% of all warnings in treatment mode
– Lane tracking
g is available 63% for both edges
g ((71% for
left only) -over the 25 mph minimum operating speed.
– Left bias: 68% of these warnings occur on left
– 94% off these
h
warnings
i
are considered
id d valid.
lid
• Lane change merge warnings:
– 94% are considered
id d valid:
lid Adjacent-lane
Adj
tl
vehicle,
hi l and
d
host vehicle is drifting over lane edge, or has initiated a
g
lane change
– 68% of these warnings occur on left side
October 20, 2010
91
Warning Events & Driver
E
Experience
i
• Forward collision warning (FCW):
– 84% of warnings to targets that were seen moving were
valid
– 2% of warnings for always-stopped targets are
considered valid
– 44% off FCWs
FCW are triggered
ti
d by
b fifixed
d objects
bj t
• Curve speed warning (CSW):
– 74% valid:
lid th
the vehicle
hi l ttraverses a significant
i ifi
t curve
– Invalid warnings for exit ramps that the driver does not
take and warnings triggered by map data error and/or its
interpretation
October 20, 2010
92
Warning Events & Driver
E
Experience
i
• Invalid warnings are inevitable
– With current technology and are particularly pronounced
for “predictive”
p
systems
y
such as FCW and CSW that
maintain an “always-available” state
October 20, 2010
93
LV Comprehensive System
Results:
Driver Behavior
• There was no effect of the integrated system
on frequency of secondary tasks.
– 2,160
2 160 sampled
l d video
id clips
li were coded
d d ffor
secondary tasks
– Younger & middle
middle-age
age drivers had more
secondary task involvement than older drivers
in their 60s (p
(p=.001)
.001)
– More secondary tasks occurred at night
(p
(p=.003)
)
October 20, 2010
94
LV Comprehensive System
Results:
Driver Behavior
• In multiple threat scenarios
scenarios, the initial
warning was generally enough to get the
attention of drivers
– Only six cases of this observed were during treatment
mode
October 20, 2010
95
LV Comprehensive System
Results:
Driver Acceptance
• Drivers believe that the integrated system
will increase their driving safety
Res
sponse Freque
ency
Overall, I think that the integrated system is going to increase
my driving safety.
25
20
15
Mean (Y) = 5.0
M
Mean
(M) = 5
5.5
5
Mean (O) = 6.1
St. Dev (Y) = 1.5
St
St. Dev
D (M) = 1
1.3
3
St. Dev (O) = 1.1
Y
Younger
Middle-aged
Older
10
5
0
1
Strongly
gy
Disagree
October 20, 2010
2
3
4
5
6
7
Strongly
gy
Agree
96
LV Comprehensive System
Results:
Driver Acceptance
• Drivers rated the integrated system well in
terms of both usefulness and satisfaction
20
2.0
BSD
Satisfactio
on
1.5
LDW
1.0
IVBSS
CSW
LCM
05
0.5
FCW
0.0
0.0
Neutral
0.5
1.0
1.5
2.0
Usefulness
October 20, 2010
97
LV Comprehensive System
Results:
Driver Acceptance
• The majority of drivers reported that they
were willing to purchase the system
Res
sponse Frequ
uency
What is the maximum amount that you would pay for the
integrated system?
50
40
30
20
10
0
Maximum Price
October 20, 2010
98
LV Lateral Control and Warnings:
Di
Driver
B
Behavior
h i
• The integrated system had a significant
effect on reducing lane departures
– Decreasing
D
i ffrom 14
14.6
6 tto 7
7.6
6 per 100 miles
il
35
Departures pe
er 100 miles
30
25
20
1
15
10
5
0
October 20, 2010
Baseline
-5
Treatment
99
LDW/LCM Example Videos
October 20, 2010
100
LV Lateral Control and Warnings:
Di
Driver
B
Behavior
h i
• Drivers 0
0.9
9 cm closer to center with the
integrated system
• An
A occupied
i d adjacent
dj
t llane causes d
drivers
i
tto
shift 10 cm away from the object/vehicle
Left of lane center
October 20, 2010
101
LV Lateral Control and Warnings:
Di
Driver
B
Behavior
h i
• Drivers use turn signals more often when
changing lanes
Percent of U
Unsignaled Lan
ne Changes
16%
14%
12%
10%
8%
Baseline
Treatment
6%
4%
2%
0%
Surface
October 20, 2010
Limited Access
102
LV Lateral Control and Warnings:
Di
Driver
A
Acceptance
t
• Not all lateral warnings were considered
necessary by drivers
Res
sponse Freque
ency
The integrated system gave me left/right hazard warnings when
I did not need one.
70
60
50
Mean = 4
4.2
2 St
St. Dev = 2
2.1
1
40
30
20
10
0
1
Strongly
Disagree
October 20, 2010
2
3
4
5
6
7
Strongly
Agree
103
LV Longitudinal Control and Warnings:
Di
Driver
B
Behavior
h i
• No effect of the integrated system on
forward conflict levels
25%
Baseline
Treatment
Percent of Even
nts
20%
15%
10%
5%
0%
‐3.25
‐2.75
‐2.25
‐1.75
‐1.25
‐0.75
‐0.25
>0
Required deceleration (m/sec2)
October 20, 2010
104
LV Longitudinal Control and Warnings:
Di
Driver
B
Behavior
h i
• No effect on the frequency of hard
hard-braking
braking
• No effect on the curve taking behavior, or
b h i
behaviors
when
h approaching
hi curves
• Close following, <1 sec HTM, increased
slightly
– 21% to 24%
• Several instances where FCWs prevented a
rear-end crash
October 20, 2010
105
FCW Example Videos
October 20, 2010
106
LV Longitudinal Control and Warnings:
Di
Driver
B
Behavior
h i
• No effect on peak sustained lateral
accelerations in curves
• No
N effect
ff t on sustained
t i d resultant
lt t acceleration
l ti
in curves
– lateral and longitudinal
• No effect on deceleration approaching a
curve
October 20, 2010
107
LV Longitudinal Control and Warnings:
Di
Driver
A
Acceptance
t
• Drivers rated the usefulness and satisfaction
of FCW lowest
– But
B t still
till neutral,
t l tto slightly
li htl ffavorable
bl
• Prominent among attributes that drivers did
nott like
lik was the
th brake
b k pulse
l for
f FCWs
FCW
October 20, 2010
108
Perceived Benefits
• Eight of 28 drivers that participated in one of
the 3 focus groups reported that the system
helped them avoid a crash
• In the individual debriefing sessions, several
additional
dditi
ld
drivers
i
stated
t t d th
the same thi
thing
October 20, 2010
109
Reactions to Warning
F
Frequency
and
dN
Nuisances
i
• Drivers noted nuisance warnings in
questionnaires, debrief interviews, and focus
groups
• Older drivers were more forgiving than
middle-aged
iddl
d or younger d
drivers
i
– Invalid rates were relatively constant across age
groups
October 20, 2010
110
Warning Modalities
• Drivers reported warnings were:
– attention-getting, not annoying
– generally
ll predictable
di t bl and
d consistent
i t t
• Drivers used the visual display to confirm
warning type, and found the display useful
• Drivers found volume control useful
• Few drivers used the mute button
• Many drivers disliked the FCW brake pulse
October 20, 2010
111
Open-Ended
Open
Ended Impressions
• What did you like most about the integrated
system? (Top three responses)
– Blind spot detection icons
– LDW drift warnings
– Greater awareness of driving habits
• What did you like least about the integrated
system? (Top three responses)
– False warnings
– Brake pulse which accompanied FCWs
– Auditory tones: too startling for some drivers
drivers, other drivers wanted
“voice warnings” instead of tones
October 20, 2010
112
Summary – LV FOT
• Drivers generally accepting of the integrated
system
• Drivers
Di
reported
t d th
thatt bli
blind-spot
d
t detection
d t ti
was the most useful and satisfying
• Secondary task activity unchanged
• Fewer lane departures,
p
, increased turn
signal use
Multiple-threat
threat warning sequences are rare
• Multiple
October 20, 2010
113
Independent Evaluation
– Light Vehicle
Preliminary
y Results
IVBSS Public Meeting
October 20, 2010
114
Presentation Outline
•
•
•
•
115
Evaluation status
Goals and objectives
y approach
Analysis
Preliminary results
– Safetyy impact
p
– Driver acceptance
– System performance
Evaluation Status
• Analysis
y currently
y underway
y
• Draft report due December 12, 2010.
116
Independent Evaluation Goals
Safety Impact
Safety Impact
• Safety benefits
• Unintended consequences
Driver Acceptance
• Ease of use • Perceived usefulness • Driving performance
• Ease of learning • Advocacy
System Performance
• Sensors • Interfaces
Sensors
• Interfaces
• Alert Logic • Robustness
117
Analysis Approach
118
Rear End
R
Road Lan
ne Change Deeparture
Pre‐Crrash Scenario
Target Pre-Crash Scenarios
Negotiating a curve/lost control
6.8%
Road edge departure/curve
3.3%
Road edge departure/straight road
10.6%
Opposite direction/no maneuver
3.9%
Drifting/same direction
1.9%
g
Turning/same direction
7.3%
Changing lanes/same direction
11.6%
Lead vehicle at constant speed
6.6%
Lead vehicle decelerating
14.1%
L d hi l t
Lead vehicle stopped
d
0%
33.9%
5% 10% 15% 20% 25% 30% 35%
Portion of Light Vehicles of Interest
2,674,000 light vehicles of interest were involved
in target pre-crash scenarios annually
(based on average 2004-2008 GES statistics)
119
Safety Impact Framework
Overall Driving
Driving Performance
Driver Inattention
Frequency of Occurrence
Driving Conflicts
Near
Crashes
Driver Response
Frequency of Occurrence
Driver Inattention
120
Projection
of
Potential
Safety
Benefits
Overall Driving Measures
• Driving Performance
–
–
–
–
Travel speed
Time headway
Lane change maneuvers
Lane keeping
• Driver
Di
IInattention
tt ti
– Secondary tasks
– Eyes-off-forward-scene
Eyes off forward scene
121
Driving Conflict Measures
• Exposure
– Rear end conflicts
• Lead vehicle decelerating/moving at slower speed
• Lead vehicle stopped
– Lateral conflicts
• Lane change
• Road departure
– Curve
C
speed
d conflicts
fli t
• Response
– Rear end conflicts: time
time-to-collision/deceleration
to collision/deceleration level
– Lateral conflicts: lateral acceleration/lane bust distance
– Curve speed conflicts: deceleration and lateral acceleration
122
Video Analysis
All Alerts
Video Available
Distraction
Steering Response
Host Vehicle Position
Crash Imminent
Eyes off Forward Scene
Host Vehicle Maneuver
Location
FCW Alerts
LDW-I /LCM Alerts
Target Type
Target Type
Target Vehicle Body Type
Target Location
Lead Vehicle Maneuver
Moving Target Vehcicle Relative Speed
Lead Vehicle Position
In path of host Vehicle
Lead Vehicle Maneuver Times
CSW Alerts
Traverse Curve
Passed Road Split
LDW-I/LDW-C
Lane Excursion Scenario
Lane Marker
Road Condition
Opposing Traffic
Time of Collision
FCW: Forward Crash Warning; LCM: Lane Change/Merge; CSW: Curve Speed Warning
LDW/I-C: Lane Departure Warning/Imminent-Cautionary
123
Preliminary Light Vehicle Results
124
Rear End Conflict Rates
Num ber of conflictts per 100 miiles
Younger
Middle‐aged
Older
5.5
5.0
4.5
4.0
3.5
3.0
B
T1
Test period
Younger: 20-30 years old
Middle-aged: 40-50 years old
Older: 60-70 years old
125
All Drivers
T2
Curve Speed Conflict Rates
Num
mber of conflictts per 100 milles
Younger
Middle‐aged
Older
2.0
1.5
1.0
0.5
00
0.0
B
T1
Test period
Younger: 20-30 years old
Middle-aged: 40-50 years old
Older: 60-70 years old
126
All Drivers
T2
Lane Change/Merge Conflict Rates
Num ber of conflictts per 100 miiles
Younger
Middle‐aged
Older
1.1
10
1.0
0.9
0.8
0.87
07
0.7
0.66
0.6
0.5
0.4
B
T1
Test period
Younger: 20-30 years old
Middle-aged: 40-50 years old
Older: 60-70 years old
127
All drivers
T2
Road Departure Conflict Rates
Num
mber of confliccts per 100 m iles
Younger
Middle‐aged
Older
2.3
22
2.2
2.1
2.0
1.9
1.8
1.7
1.6
15
1.5
B
T1
Test period
Younger: 20-30 years old
Middle-aged: 40-50 years old
Older: 60-70 years old
128
All Drivers
T2
Conflict Rates
Num ber of conflictts per 100 milles
Younger
Middle‐aged
Older
10.0
98
9.8
9.6
9.4
9.2
9.0
8.8
8.6
8.4
8.2
8.0
B
T1
Test period
Younger: 20-30 years old
Middle-aged: 40-50 years old
Older: 60-70 years old
129
All Drivers
T2
Driver Acceptance – Demographic Breakdown
Education Level
• Higher education level
• Higher
g
education level
Lower understanding of warnings
More annoyance
y
with system
y
Years of Driving Experience
• More experience
p
Less annoyance
y
with nuisance warnings
g
• More experience
System made driving easier
130
Driver Acceptance – Driver Experience
B
Breakdown
kd
•
More alerts
Better understanding of warnings
•
Shorter average trip length
•
Less freeway driving
System made driving easier
System more consistent
Fewer nuisance warnings
•
More nighttime driving
Less distracted by warnings
Better understanding
g of warnings
g
All results significant based on paired t-test
131
Driver Acceptance – Driver Experience Breakdown
Responses to: “The nuisance warnings were not annoying”
Lower alert rates
Strongly
Agree
Higher alert rates
g
7
6
5
4
3
2
Strongly
Di
Disagree
P=0.01
P=0.04
P=0.26
1
0
FCW
132
P=0.08
CSW
Side Hazard
Drift
Alert Rates
Baseline
Treatment
12
11.2
Ale
erts per 100 miles
10
8
5.8
6
4
2
0.4 0.4
0.4
0.4
0.7 0.6
1.3 1.1
0
FCW
CSW
LCM
Alert Type
133
LDW‐I
LDW‐C
Alert Rates
# of Alerts per 100 mi
•
•
Baseline
Treatment
All Alerts
15.4
8.4
46%
Imminent Drift
1.3
1.1
14%
Cautionary Drift
12.7
5.9
54%
All alerts/cautionary drifts significant for all age/gender groups
Females showed a reduction in imminent drift warnings
All results significant based on paired t-test
134
% Reduction
Preliminary Secondary Task Results – Most
C
Common
T
Tasks
k
Secondary tasks present in 57% of alerts
lk
/l k
at passengers
Talking to/looking 19%
Grooming
9%
Talking on/listening to phone
7%
Looking to the side/outside car
g
/
6%
Adjusting controls
4%
Text messaging
3%
Smoking/lighting cigarettes
2%
Eating
2%
Reaching for object in vehicle
2%
Reading cell phone
2%
0%
5%
10%
15%
Porportion of Alerts
(Based on video analysis of 14,860 alerts)
135
20%
25%
Preliminary Secondary Task Results - Most
C
Common
S
Secondary
d
Tasks
T k by
b Age
A Group
G
Olderr
Talking on/listening to phone
6%
Looking outside car
Looking outside car
8%
Grooming
9%
Middle‐Ageed
Talking to/looking at passengers
19%
Talking on/listening to phone
7%
Grooming
8%
Looking outside car
8%
Talking to/looking at passengers
17%
Younger
Texting
6%
Grooming
9%
Talking on/listening to phone
10%
Talking to/looking at passengers
23%
0%
5%
10%
15%
Porportion of Alerts
(Based on video analysis of 14,860 alerts)
136
20%
25%
Preliminary Eyes-Off-Forward Scene Results
Po
oportion of Alertss with Eyes‐Off‐Forward SScene
10%
9%
9%
8%
8%
7%
6%
6%
5%
4%
3%
2%
1%
0%
Younger
Middle‐Aged
Most common for LDW-C alerts for all age groups
(Based on video analysis of 14,860 alerts)
137
Older
Questions
138
Field Operational Test
Data System Overview
Scott Bogard
UMTRI
Contributors: J. Sayer, D. Funkhouser,
A. Blankespoor, S. Bao
M. L. Buonarosa, D. LeBlanc
Primary IVBSS FOT
D t b
Databases
•Design optimized for:
o Back-up
o Query
Q
efficiency
ffi i
o Data loading
•Views relate Radar and Raw to FOT
•Read onlyy
October 20, 2010
140
UMTRI Data Guidelines
– All data belongs in a database
• This includes system definition and collected or
derived data
– All databases should be fully normalized
• Data should not be duplicated
• Data must have common ‘keys’ which form a unique
entry in a database table
– Driver,
Driver Trip
Trip, Time (all integer numbers)
– Additional keys: Index, Unique Id numbers
– If designed
g
correctly
y a relational database
architecture is very efficient and scalable
October 20, 2010
141
IVBSS Data Structures
• Database tables (LV 93/741GB; HT
97/1.26TB)
– Time
Ti
history
hi t
• Time sequenced data typically at 10 Hz
• More complex for radar which has multiple targets at
any given time creating a one-to-many relationship
• Transition: logged upon change in state (PRNDL)
October 20, 2010
142
IVBSS Data Structures and
A hi
Archive
– Database tables
• Summary or Event
– aggregated
gg g
data over a p
period of time
• Histograms—frequency distributions
• Reference—mapping between a value and it’s
meaning
i
– 4.5 B records in largest HT table (radar)
– 1.2
1 2 B records
d ffor LV
• Video—time stamped, mpeg4 compressed in a
bit
bitmap
fformatt (LV 1
1.09
09 TB
TB; HT 2
2.19
19 TB)
October 20, 2010
143
IVBSS Primary Content
• LV FOT and HT FOT
– Vehicle—state variables and driver inputs
– Warning
W i technology
t h l
• Radar and lane tracking camera
– Diagnostic
• Faults; heart beats; histograms; odometer; last
values
– GPS
– Summary
October 20, 2010
144
IVBSS Secondary Content
– LV and HT (256 and 238 tables)
•
•
•
•
Longitudinal/Lateral conflict metrics
Traffic surrogates from radar tracks
Signal conditioning
Events: curve, straight, moving, stopped, road type,
dark, braking, following, lane change…
– Supplemental data sources
• Highway Performance Monitoring Sys.
– Intersections and nodes, traffic count, signaling….
• Weather
October 20, 2010
145
Analysis tools and video
content
t t
• Most ‘off
off-the-shelf
the shelf’ analysis tools can
interact with a database… import and export
• Most
M t tools
t l have
h
a programming
i llanguage
for iterative analysis
• All UMTRI derived tools interface directly
with the database and video files for
efficient/content rich analysis and review of
events
October 20, 2010
146
Laptop Demonstration
• Live demonstration showing the evolution of
SQL queries to find exciting events of
braking for slowing vehicles
vehicles….includes
includes the
UMTRI viewer.
October 20, 2010
147
Access to Field Operational Test Data
and Future Uses
Dave LeBlanc / Jim Sayer
UMTRI
Sharing IVBSS FOT Data
Goals:
 Accelerate successful deployment of crash avoidance
technologies


Light vehicle
Heavy vehicle
 Maintain the data archive
 No support funding is in place - costs of data upkeep are shared
by
yU
UMTRI & outside investigators
g
Constraints:
 Protecting rights & privacy of test participants
 Protecting commercial interests of industry partners
 Existing research interests
Focus of Discussion
 Today: Addressing themes of IVBSS
research:




Crash warning technologies
Driver interaction and adaptations to
technology
Driver acceptance
G
Generic
i vehicle-based
hi l b
d performance
f
capabilities and limitations
 Not today: Other research themes being
pursued with this & other data sets
Application of FOT Data Sets
t Subsequent
to
S b
tR
Research:
h P
Partners
t
Industry Partners







GM
Ford & Volvo Car
H d
Honda
Nissan
Delphi
Eaton
Bosch
Gov’t/Research Groups











US DOT
US DOE

US A
Army
National Academy
of Sciences/TRB
Alzheimer's Ass’n
SAIC/VTTI
U. Iowa
U. Minnesota
Penn State Univ.
Iowa State
George Marshall
Univ.
Chemnitz Inst.
Inst
of Technology
Data Resources
 IVBSS FOT data
 Four other FOT data sets
 Roadway
y network data sets
 Weather data
 Crash data – UMTRI Transportation Data
Center
 Other data sets can be brought
g in for
specific purposes
Examples*
Examples
*: Simple Access to Data
 Driver lane-keeping data samples
 Delivered or remote access
 Driver head motion during lane changes
 Video review by outside or UMTRI staff,
staff remote
or on-site
 Frequency
q
y and kinematics of lead vehicle
turning scenarios (potential false alerts)
 Age
g differences in driving
g performance
p
 Correlating travel patterns & conflict
frequency
* Actual and hypothetical examples included
Examples: Fusing Roadway Data
 Speeding (relative to posted speed) and
lane change behavior
 Driving behavior at signalized
intersections
 Urban driving behavior vs. suburban
 Lateral acceleration on freeway ramps
Examples: ResearchResearch-Level Use of
D t
Data
 V2V: Estimates of possible reduction in
unwanted alerts due to V2V information
 Driver acceptance as a function of alert
experience and demographics
 Driver style classification and alert design
UMTRI Work Tasks
Naturalistic data-related collaborations often
involve:
 ‘Standard’ data processing and packaging
tasks
 Customized data work
 Fusing other data sets
 Research collaboration-level work
Process:
Match the Problem and the Data
 Investigator discusses specific interest
with UMTRI. For technical issues:




Data dictionary is shared & nature of the data
products discussed
Sample data may be shared
Video/numerical data is sometimes reviewed
Investigators often discover new possibilities
for addressing their issues
 Data richness (data fusion),
 Ability to isolate (querying),
 Statistical techniques.
q
Process:
Identify Deliverables & Tasks
 Identify data products, data access, and
tools



Data products: raw data, pointers into data,
smoothed data, fused data, aggregated data,
video possibly with overlays
Access to data: delivered to investigator,
remote
t access, on-site
it access
Tools to be provided and/or developed (e.g.,
viewing/coding video)
 UMTRI work and/or research tasks are
identified
Process:
Contractual/IP
 Nondisclosure agreements for early
conversations, if investigator prefers
 Agreement type usually depends on the nature of
the technical work

P.O.
P
O
 Research agreement,
 Research agreement with restricted rights
 All the above include a data sharing agreement
addressing the investigator’s access & use of
data:

Limited time, for specific research purpose, and with a
limited number of people
 Data security
sec rit (sometimes)
Process:
Protecting Individual Identities
 Sometimes investigator gains access to data that
allows individual participants to be identified

Driver face video
 GPS coordinates for a set of complete trips (allows
residence to be identified)
 If so:
 Another agreement with UMTRI is usually required,
restricting access within the organization
 Investigator may need approval from a certified board
for IRB
 These issues are often avoided by ‘de-identifying’
the data
Process:
P t t IVBSS iindustry
Protect
d t partners
t
 Commercial interests of partners
 Minimize an outside investigators' ability to observe
sensitive information about details of sensor capability,
algorithmic approaches, or calibration values.
 Techniques include:
 Smoothing of signals
 Non-random sampling of episodes to mask sensitive
details of performance
 Avoiding coupling of algorithm inputs & outputs
 Any techniques used are discussed with potential
investigator.
Contacts
 Dave LeBlanc
 leblanc@umich.edu
 (734) 936-1063
 Jim Sayer
 jimsayer@umich.edu
 (734) 764-4159