Upload
transportforum-vti
View
277
Download
0
Tags:
Embed Size (px)
DESCRIPTION
Citation preview
Johan Engström, Volvo Technology
Transportforum, Linköping 2011-01-13
Hur leder kritiska händelser till allvarliga olyckor?
Accidents and incidents
9. F
ata
l cra
sh
7. P
rope
rty
dam
age
only
6. L
ow-g
ph
ysic
al c
ont
act
5. T
ire s
trik
e
4. N
ear-
cras
h
3. C
rash
-re
leva
nt c
onf
lict
2. P
roxi
mity
con
flict
1. N
on-c
onf
lict
0. Ir
rele
vant
ev
ent
8. A
ir-ba
g d
eplo
ymen
t an
d/o
r in
jury
Normal driving
Crash-Relevant EventCategory
Incid
ent T
ype
Crash
Type
Severity
Type
9. F
ata
l cra
sh
7. P
rope
rty
dam
age
only
6. L
ow-g
ph
ysic
al c
ont
act
5. T
ire s
trik
e
4. N
ear-
cras
h
3. C
rash
-re
leva
nt c
onf
lict
2. P
roxi
mity
con
flict
1. N
on-c
onf
lict
0. Ir
rele
vant
ev
ent
8. A
ir-ba
g d
eplo
ymen
t an
d/o
r in
jury
Normal driving
Crash-Relevant EventCategory
Incid
ent T
ype
Crash
Type
Severity
Type
p
Intro
• Key questions:– How do normal driving develop into critical events?– How do critical events develop into crashes?
• Traditionally: Limited data to answer these questions (crash reconstruction, subjective reports)
• Today– Naturalistic data with real crashes and near-crashes
captured on video– Need to know what to look for – need for an
accident model for micro-level analysis of the pre-crash phase
Anticipatory selection in driving
• Driving involves proactive selection of relevant information and actions
• Based on expectations on how an upcoming situation will develop
• Can be conceptualised in terms of schemata
Schemata
• Functional representations of actions, or action sequences making up a task
• Embody ”implicit” expectations
• Learned with experience
• Overlearned schemata often selected and excecuted automatically without conscious awareness, even at task level (”zombie behaviour”)
• Schema selection = attention selection
Turn right at intersection
Look left for carsSlow down Turn right
Task context schema
Basicschemata
Attention/action selection model
Schemata
Environment
Sensing Actuation
SchemataBasic
Task context
Cognitive control
SchemataTask context
Top-down
Bottom-up
• Task context and basic schemata
• Related schemata may compete for activation
• Schemata selected top-down (proactively) and/or bottom-up (reactively)
• Two types of top-down selection
– Context-driven (automatic, unconscious, inflexible)
– Cognitive control (effortful, conscious
Critical situations
Critical situations occur due to a conflict between the proactively selected schema and how the actual situation develops
Schemata
Environment
Sensing Actuation
SchemataLook for pedestrians
Task context
Cognitive control
SchemataDrive through intersection
Top-down
• Possible reasons:– Infrequent event
– Misleading traffic/infrastructure configuration
– Attentional ”capture” of competing schema (distraction)
– Unfamiliarity
– Working memory load
– ….
How do critical situations lead to crashes?
• Possible reasons:– Insufficient time to react– Stimulus outside field of view
(e.g. due to off-road glance)– Blink– Occlusion– Change blindness– Glare– Low stimulus saliency– Microsleep– ….
A critical situation will lead to a crash if last-second, reactive schema selection fails
Schemata
Environment
Sensing Actuation
SchemataBrake!
Task context
Cognitive control
SchemataDrive through intersection
Top-down
A general accident model
Normal driving
Critical event
Crash
Schema
Situation
Conflict ?Successful bottom-up selection?
No
Nearcrash
time
Reactive barrierProactive barrier
Yes
No
Yes
Real world example 1: Car-bicycle crashes at intersections and roundabouts (Summala and Räsänen, 2000)
Interpretation – Summala study
Driver’s schema: Look for cars to the left
Bottom-up selection: None since the bicyclist appears outside the field of view
Actual situationBicyclist approaching from the right
Normal driving
Critical event
CrashConflict ? No
Reactive barrierProactive barrier
Yes
Real-world example 2: Rear-end crash in the 100-car study
Narrative (by VTTI analyst): Subject driver is approaching a right turn at an intersection. The lead vehicle briefly stops at stop sign, then moves forward as if completing the turn. As the subject driver looks out his left window to check traffic, the lead vehicle stops again. The subject vehicle hits the lead vehicle in the rear. Inopportune glance.
Event 8633
Brake
GazeSpeed
Impact
Acceleration
Interpretation – 100 car rear end crash (event 8633)
Driver’s schema: Lead vehicle will continue to turn -> OK to look left to check traffic
Actual situationLead vehicle stops again
Bottom-up selection: None since the lead vehicle braking occurs outside the field of view
Normal driving
Critical event
CrashConflict ? No
Reactive barrierProactive barrier
Yes
Aggregate analysis, 8 rear-and crashes
Homogenousmechanisms!
Aggregate analysis, 8 near crashes
More heterogenousmechanisms!
Lee et al. (2007) quantitative analysis of 100-car study rear end events (same data)
Eyes-off-road is what distinguishes the crashes
TTC similar between crashes and near crashes
Some implications
• Hard to define ”crash-relevant events” in pure kinematic terms
• Data on driver’s proactive schema selection ideally requires both video and subjective report
• However, proactive selection often unconscious -> drivers may not be able to report! (aneqdotal data from Hanowski)
Conclusions
• A critical situation occur due to a mismatch between the proactively selected schema and how the actual situation develops -> breaking proactive barrier
• A critical event will lead to a crash if last-second, reactive schema selection fails -> breaking reactive barrier
• Accident/incident analysis at micro level should find out key mechanisms behind breaking of proactive and reactive barriers
• Requires combination of naturalistic and laboratory (simulator) data
• Future goal: Coherent taxonomy and classification scheme