Fundamental probabilistic analysis analysis Fundamental - - PowerPoint PPT Presentation

1

Fundamental probabilistic Fundamental probabilistic analysis analysis

• n effectiveness
• n effectiveness
• f safety
• f safety-
• presentation type

presentation type

• n safe driving support system
• n safe driving support system

○ ○K. OKABE, M. KAMATA

• K. OKABE, M. KAMATA (Tokyo

(Tokyo Univ

• Univ. JPN)

. JPN)

• T. HIRAOKA, H. KUMAMOTO (Kyoto
• T. HIRAOKA, H. KUMAMOTO (Kyoto Univ
• Univ. JPN)

. JPN)

2

Current Situation Current Situation

• Traffic road accidents in Japan

Traffic road accidents in Japan

• About 10,000 people are killed (decreasing)

About 10,000 people are killed (decreasing)

• About 1,000,000 people are injured (increasing)

About 1,000,000 people are injured (increasing)

AHS( AHS(Advanced Cruise

Advanced Cruise-

• Assist Highway System

Assist Highway System)

)

• Seven support services

Seven support services

• Necessity of quantitative assessments

Necessity of quantitative assessments

⇒

3

Source: Ministry of Land, Infrastructure and Transport

4

Seven Services of AHS Seven Services of AHS

• Prevention of collision with forward obstacles

Prevention of collision with forward obstacles

• Prevention of overshooting on curve

Prevention of overshooting on curve

• Prevention of lane departure

Prevention of lane departure

• Prevention of crossing collisions

Prevention of crossing collisions

• Prevention of right turn collisions

Prevention of right turn collisions

• Prevention of collisions with pedestrians

Prevention of collisions with pedestrians crossing streets crossing streets

• Road surface condition information for

Road surface condition information for maintaining headway,etc maintaining headway,etc

5

Safety Analysis Safety Analysis

• Safety alarm for forward obstacle curve collision

Safety alarm for forward obstacle curve collision

• Road

Road-

• Vehicle communication system

Vehicle communication system

STOP STOP

Wireless

6

Discussion Point Discussion Point

• Configuration of safety monitoring system

Configuration of safety monitoring system

• Fault warning type

Fault warning type & & Safety presentation type Safety presentation type

Fault Warning Safety Presentation

Danger ! Safe !

STOP GO

Which is more appropriate?

7

Comparative Study Comparative Study

• Evaluation of implementation effect

Evaluation of implementation effect

• Possibility of decreasing accidents (normal operation)

Possibility of decreasing accidents (normal operation)

• Scarcity of

Scarcity of increasing accidents (abnormal operation) increasing accidents (abnormal operation)

Fault Warning Safety Presentation

Danger ! Safe

STOP GO

=> Probabilities of fail dangerous failure

Obstacle ! No obstacle

8

Accident Outbreak Accident Outbreak

• Definition of the process

Definition of the process

A A: : Avoidance Action Avoidance Action

• Ex. stopping, changing lane

A A: : No Avoidance Action No Avoidance Action

• Ex. keeping on driving

D D: : Presence of Dangerous Relation Presence of Dangerous Relation

• Def. situation where A surly

cause accident

9

Fail Dangerous Failure Fail Dangerous Failure

• Fault warning

Fault warning

• Sensor fails to detect a danger relation

Sensor fails to detect a danger relation D D

• Sensor succeeds in

Sensor succeeds in detecting detecting D D but fails to send a message but fails to send a message

• Safety presentation

Safety presentation

• Sensor

Sensor fails to detect fails to detect D D and sends and sends false message false message

Fault Warning Safety Presentation

Danger ! Safe !

10

Fail Dangerous Probability Fail Dangerous Probability

• Reliabilities

Reliabilities

• Sensor detection

Sensor detection

• Communication

Communication

• Fault warning

Fault warning

• Sensor fails to detect a danger relation

Sensor fails to detect a danger relation D D

• Sensor succeeds in

Sensor succeeds in detecting detecting D D but fails to send a message but fails to send a message

• Safety presentation

Safety presentation

• Sensor

Sensor fails to detect fails to detect D D and sends and sends false message false message

QuickTime?and a Photo - JPEG decompressor are needed to see this picture.

QuickTime?and a Photo - JPEG decompressor are needed to see this picture.

QuickTime?and a Photo - JPEG decompressor are needed to see this picture. QuickTime?and a Photo - JPEG decompressor are needed to see this picture.

11

Study Result #1 Study Result #1

• Evaluation of implementation effect

Evaluation of implementation effect

• Possibility of decreasing accidents (normal operation)

Possibility of decreasing accidents (normal operation)

• Scarcity of

Scarcity of increasing accidents (abnormal operation) increasing accidents (abnormal operation)

Safety Presentation

=> Probabilities of fail dangerous failure

Fault Warning

Danger ! Safe

STOP GO

Obstacle ! No obstacle

is more appropriate

12

Comparative Study Comparative Study

• Evaluation of implementation effect

Evaluation of implementation effect

• Possibility of decreasing accidents (normal operation)

Possibility of decreasing accidents (normal operation)

• Scarcity of

Scarcity of increasing accidents (abnormal operation) increasing accidents (abnormal operation)

Fault Warning Safety Presentation

Danger ! Safe

STOP GO

=>Estimated Accident Probability after Implementation

Obstacle ! No obstacle

13

Accident process Accident process

• Event

Event Tree Tree Accident No accident

Driver action

No accident

Dangerous relation

D D: Non : Non-

• Presence of Dangerous Relation

Presence of Dangerous Relation D D: Presence of Dangerous Relation : Presence of Dangerous Relation A A: Avoidance Action : Avoidance Action A A: No Avoidance Action : No Avoidance Action

Message Message

： ：How does it change ?

How does it change ?

D D D D A A A A

14

Effect of Message Effect of Message

Message Dangerous relation Action STOP STOP zZZ

Success

D D

Fault warning type

Fail

Message Dangerous relation Action Go Go zZZ

Success

D D

A A

Fail

Safety presentation type

A A A A A A

Accident

15

Effect of safety device Effect of safety device

• Evaluation method

Evaluation method

• Experimental approach (case by case: bottom up)

Experimental approach (case by case: bottom up)

• e.g. driving simulator base

e.g. driving simulator base

• Theoretical

Theoretical approach approach (general purpose: top down) (general purpose: top down)

• e.g. concept base

e.g. concept base

Accident No accident Driver action Dangerous relation Message Message ： ：How does it change ?

How does it change ?

D D A A A A

Warning

DANGER

16

Topics Topics

• Theoretical

Theoretical Approach: Cognitive Driver Model Approach: Cognitive Driver Model

• Concept: risk homeostasis hypothesis

Concept: risk homeostasis hypothesis

• Our proposed model:

Our proposed model: maximum acceptable risk model maximum acceptable risk model

17

Risk Homeostasis Theory Risk Homeostasis Theory

• Outline of the risk homeostasis theory

Outline of the risk homeostasis theory

• A driver behaves based on a target level of risk

A driver behaves based on a target level of risk

• An accident rate fluctuates around a stable mean

An accident rate fluctuates around a stable mean

=> Risk Homeostasis

Wild, G. J. S: Target Risk, PDE Publications, 1994

18

Target Level of Risk Perceived Level of Risk Decisions Taken Actions upon Vehicle Controls Vehicle Response

Comparator, Summing Point

+

• Time Lagged Feedback

Information Intake

Reference: Wild, G. J. S, The theory of Risk Homeostasis: Implications for Safety and Health, Risk Analysis, 1982

Task Model (Target Risk Model) Task Model (Target Risk Model)

Conditions

• A driver behaves based on a target level of risk

A driver behaves based on a target level of risk

19

Target Level of Risk Perceived Level of Risk Desired Adjustment: | a-b-c| ≈ 0 Adjustment Action Resulting Accident Loss

+

• Time Lagged Feedback

a b

Protection Features

Estimates

• f Protection Effect

c

• => Before ≈ After

Risk Homeostasis Model Risk Homeostasis Model

• An accident rate fluctuates around a stable mean

An accident rate fluctuates around a stable mean

STOP STOP

20

Adjustment Action Adjustment Action Adjustment Action Individual Target Level of Risk Individual Perceived Level of Risk Adjustment Action

Accident Rate Accident Rate

Time Accident Rate

A stable mean value

Accident Rate Accident Rate

21

Argument for more than a decade Argument for more than a decade

• Fruitless argument

Fruitless argument

• Adams, 1981: The efficacy of seat belt legislation

Adams, 1981: The efficacy of seat belt legislation ---

• is one of evidence (by Wilde: the author)

is one of evidence (by Wilde: the author)

• Grime, 1979: A review of research on the protection

Grime, 1979: A review of research on the protection ---

• is one of contrary evidence (by

is one of contrary evidence (by MacKenna MacKenna, 1982) , 1982) => Wiled, 1984; => Wiled, 1984; MacKenna MacKenna, 1982 is not sufficient analysis , 1982 is not sufficient analysis => Shannon, 1986: Road accident data => Shannon, 1986: Road accident data ---

• gave new contrary evidence

gave new contrary evidence => Evans 1986: Risk Homeostasis theory and traffic => Evans 1986: Risk Homeostasis theory and traffic ---

• gave some new contrary evidences and denied

gave some new contrary evidences and denied => Wiled => Wiled, 1986; Evans, 1986 does not mean contrary evidence. 1986; Evans, 1986 does not mean contrary evidence. f field experiments are sufficient analyses

ield experiments are sufficient analyses More arguments were yields

22

y = 7E-07x - 96949 R2 = 0.8741 300000 320000 340000 360000 380000 400000 420000 440000 460000 480000 5E+11 5.5E+11 6E+11 6.5E+11 7E+11 7.5E+11 8E+11

Total travel distance per year Total number of accidents per year

Strong correlation Strong correlation

23

0.45 0.47 0.49 0.51 0.53 0.55 0.57 0.59 0.61 0.63 0.65

1989 1990 1991 1992 1993 1994 1995 1996 1997 1998 1999 2000 Year 3.0E-07 3.1E-07 3.2E-07 3.3E-07 3.4E-07 3.5E-07 3.6E-07 3.7E-07 3.8E-07 3.9E-07 4.0E-07 Accident rate ( 蠶)

• Accidents in intersections with traffic signals

Accidents in intersections with traffic signals

• Accident rate:

Accident rate: accidents / ( travel distance accidents / ( travel distance × × density of signal) density of signal)

Transition of accident rate Transition of accident rate

24

Topics Topics

• Cognitive Driver Model

Cognitive Driver Model

• Concept: risk homeostasis hypothesis

Concept: risk homeostasis hypothesis

• Our proposed model:

Our proposed model: maximum acceptable risk model maximum acceptable risk model

• based on Target Risk Model

based on Target Risk Model

• Risk Homeostasis => ?: never concluded

Risk Homeostasis => ?: never concluded

=> Risk Compensation would be preferable => Risk Compensation would be preferable

25

Target Level of Risk Perceived Level of Risk Decisions Taken Vehicle Handling Vehicle Response Comparator, Summing Point +

• Time Lagged Feedback

Information Intake

Target Risk Model Target Risk Model

Target Risk Model shows the mechanism of Risk Compensation

Keep favorite velocity or distance Keep favorite condition

26

Target Level of Risk Perceived Level of Risk Desired Adjustment: | a-b-c| ≈ 0 Adjustment Action Resulting Accident Loss

+

• Time Lagged Feedback

a b

Protection Features

Estimates

• f Protection Effect

c

• => Before ≈ After

Risk Homeostasis: unconsciousness Risk Homeostasis: unconsciousness

• An accident rate fluctuates around a stable mean

An accident rate fluctuates around a stable mean

STOP STOP

Need not to be focal in the person’s conscious awareness

27

Target Level of Risk Perceived Level of Risk Decisions Taken Vehicle Handling Vehicle Response

Comparator, Summing Point

+

• Time Lagged Feedback

Information Intake

Need not to be focal? Need not to be focal?

Consciousness Consciousness

Target Risk Model shows the mechanism of Risk Compensation

Need not to be focal in the person’s conscious awareness

28

Perception Based Driver Model Perception Based Driver Model

• Perception result & acceptable risk

Perception result & acceptable risk

• Perception results

Perception results: Qualitative classificatory criteria : Qualitative classificatory criteria

• Acceptable risk

Acceptable risk: Quantitative assessment criterion : Quantitative assessment criterion

Danger Relation Danger Relation Perception results Perception results Acceptable Risk Acceptable Risk action action Driver model Perception Judgment Action（operation）

Situation classification Situation classification Situation assessment Situation assessment

29

Perception Result ( Perception Result (Perception Representation

Perception Representation)

)

• Three perception results

Three perception results

• Yes (

Yes (y y) ): : I I’ ’m in danger m in danger

• Ex. traffic signal is red
• No (

No (n n) ): : I I’ ’m not in danger m not in danger

• Ex. traffic signal is blue
• Unknown (

Unknown (u u) ): : I can I can’ ’t decide between t decide between Yes Yes & & No No

• Ex. traffic signal is yellow

Action Dangerous Relation

Perception Result

Driver Action

Dangerous Relation

y y n n u u

A A A A D D D D D D D D

A A

A A A A A A

A A A A

30

Acceptable Risk Acceptable Risk

• Subjective driver action

Subjective driver action

• accident occurrence

accident occurrence < < Acceptable Risk Acceptable Risk level level

• utility maximization & cost minimization

utility maximization & cost minimization accident occurrence accident occurrence = = maximum maximum of

• f Acceptable Risk

Acceptable Risk

• Objective driver action

Objective driver action

• accident probability

accident probability = = constant constant

Risk

Cost Utility

QuickTime?and a Photo - JPEG decompressor are needed to see this picture.

QuickTime?and a Photo - JPEG decompressor are needed to see this picture.

QuickTime?and a Photo - JPEG decompressor are needed to see this picture.

31 1 C 1 C unacceptable

acceptable acceptable

Probability of Danger Probability of Danger D D Non Non-

• avoidance probability

avoidance probability A A

Maximum Acceptable Risk Model Maximum Acceptable Risk Model

• Accident probability :

Accident probability :

Utility maximization, Cost minimization

QuickTime?and a Photo - JPEG decompressor are needed to see this picture.

QuickTime?and a Photo - JPEG decompressor are needed to see this picture.

QuickTime?and a Photo - JPEG decompressor are needed to see this picture.

QuickTime?and a Photo - JPEG decompressor are needed to see this picture.

32 1 1 C C

acceptable acceptable

Probability of Danger Probability of Danger D D Non Non-

• avoidance probability

avoidance probability A A

Perception & Action Perception & Action

• Performance of perception

Performance of perception

• Accurate perception cuts of

Accurate perception cuts of unnecessary unnecessary avoidance avoidance => decrease travel time => decrease travel time

QuickTime?and a Photo - JPEG decompressor are needed to see this picture.

High Accuracy High Accuracy

33

Unnecessary avoidance Unnecessary avoidance

Pr{D,A|S}

Pr{D|u}=Pr{D}=R

Pr{D|n}=C

0.2 0.4 0.6 0.8 1 1.2 1.E-08 1.E-07 1.E-06 1.E-05 1.E-04 1.E-03 1.E-02 1.E-01 1.E+00

S=n≡n* S=y≡y* S=u

33

Probability of Danger Probability of Danger D D Unnecessary avoidance action Unnecessary avoidance action

34

Driver Dependence on System Driver Dependence on System

• Drivers depend completely on the message

Drivers depend completely on the message

• Shift attention from

Shift attention from D D to message to message

Warning

DANGER

Message Dangerous relation Perception result

STOP STOP

zZZ

Success

Fail

D D D D

e.g. Fault warning type

35

Driver Dependence on System Driver Dependence on System

Message Dangerous relation Perception result STOP STOP zZZ Success Fail D D

Fault warning type

Message Dangerous relation Perception result Go Go zZZ Success D D Fail

Safety presentation type

• r
• r u

u

(wishful assumption)

36

Accident Probability Accident Probability

Message Dangerous relation Action STOP STOP zZZ

Success

D D

Fault warning type

Fail

Message Dangerous relation Action Go Go zZZ

Success

D D

A A

Fail

Safety presentation type A A A A A A

QuickTime?and a Photo - JPEG decompressor are needed to see this picture.

QuickTime?and a Photo - JPEG decompressor are needed to see this picture.

QuickTime?and a Photo - JPEG decompressor are needed to see this picture.

QuickTime?and a Photo - JPEG decompressor are needed to see this picture.

Accident

>

37

Study Result #2 Study Result #2

• Evaluation of implementation effect

Evaluation of implementation effect

• Possibility of decreasing accidents (normal operation)

Possibility of decreasing accidents (normal operation)

• Scarcity of

Scarcity of increasing accidents (abnormal operation) increasing accidents (abnormal operation)

Safety Presentation

=> Accident Probability

Fault Warning

Danger ! Safe

STOP GO

Obstacle ! No obstacle

is more appropriate

38

Thank you for your attention Thank you for your attention

• Contact address.

Contact address.

• kabe@sl
• kabe@sl.t.

.t.u u-

• tokyo

tokyo.ac. .ac.jp jp

39

Result of Safety Assessment Result of Safety Assessment

• Devices may fail to alarm

Devices may fail to alarm

• Current alarm reliability is enough ?

Current alarm reliability is enough ?

• Available reliability : 90%

Available reliability : 90% – – 95% 95%

• How many accidents may be cut down?

How many accidents may be cut down? => Not sufficient, but can reduce accident => Depend on drivers’ reliability => Up to drivers dependency on safety device (50% cut off is possible)

40

Implementation Effect (Case 1) Implementation Effect (Case 1)

• All drivers depend on safety device

All drivers depend on safety device

• Alarm reliability

Alarm reliability r r: 95% , Driver reliability 1 : 95% , Driver reliability 1-

• P

P : 90% : 90%

after after

Accident Probability Accident Probability

95% down 45%

N N： ：

up

STOP STOP

zZZ

No problem

zZZZZ

Y Y ： ：

C 0.5C

41

Implementation Effect (Case 2) Implementation Effect (Case 2)

• Some drivers depend on

Some drivers depend on

• Alarm reliability

Alarm reliability r r: 95%, Driver reliability 1 : 95%, Driver reliability 1-

• P

P : 90% : 90%

• Dependence ratio

Dependence ratio d d : 10 : 10-

• 1

1

after after

Accident Probability Accident Probability

95% 4.5 4.5% %

Dependence of all drivers : 45% Dependence of all drivers : 45%

down up

about

STOP STOP

zZZ

No problem For real ?

0.1C

d: 0.1

42

43

Required Reliability Required Reliability

• Alarm reliability:

Alarm reliability: r r

Driver action Unsafe relation

Perception result

A D D A A

y

n

u

Alarm

D

Unsafe relation

D Perception result

n r Y ⇒

r N − ⇒1

y

Driver action

A⇒C A⇒1 after

P C P r − ≅ − − ≥ 1 1 1

Driver reliability

P

1 ) 1 ( × − + × r C r

≥

∑

∑

44

Conclusions Conclusions

• Perception based driver model

Perception based driver model

• Required reliability for the alarm

Required reliability for the alarm

• Implementation effect of the safety devices

Implementation effect of the safety devices

• Importance of HI

Importance of HI