Objective Research Problem Modelling Assumptions Risk Definition ...
PROBABILISTIC DRIVING RISK ASSESSMENT APPROACH .
Freddy A. Mullakkal-Babu, Meng Wang, Bart van Arem, Riender Happee
Research Problem Surrogate safety measures are inadequate for safety assessment of mixed traffic fleet including automated vehicles:
• • •
Defined not specific to a vehicle type, considering its manoeuvre
Objective
variability
•
Limited consideration for collision severity
To specify an anticipative and probabilistic risk assessment approach that is applicable to on-road mixed vehicle interactions
•
Discontinuous in two-dimensional vehicle encounter
To derive a risk measure that can be objectively interpreted and compared
Risk Definition 𝑹𝑰𝑺𝑲 𝑴𝑬𝑨𝑺𝑼𝑹𝑬𝒔,𝒄 = 𝑷(𝑪𝒐𝒍𝒍𝒊𝒔𝒊𝒐𝒏
𝑻 𝒄,𝒋
⋅ 𝑴𝒄 ⋅ 𝒆
𝒌𝟐 ⋅|𝒗𝒄 (𝒕)|⋅𝐜𝐨𝐬 𝜽𝒄,𝒋
⋅ 𝑴𝒔 ⋅ 𝒆
Neighbour vehicle: C
−𝒌𝟐 ⋅|𝒗𝒔 (𝒕)|⋅𝐜𝐨𝐬 𝜽𝒔,𝒄
Subject vehicle: S cCollision probability
Collision severity
𝑥 𝑗 , 𝑦 𝑗 : future road position of S
𝑅𝐼𝑆𝐾 𝑀𝐸𝐴𝑆𝑈𝑅𝐸𝑠,𝑐 is the collision risk measured by the subject
Modelling Assumptions
vehicle S due to a neighbouring vehicle C. In a multi-vehicle
•
encounter, the risk measured by S is the sum of risk contributed by individual vehicles involved in the encounter.
Collision severity increases with instantaneous vehicle velocity and physical mass of interacting vehicles
•
The probability distributions of lateral and longitudinal acceleration of a vehicle type can be estimated
𝑀𝑐 , 𝑀𝑠 : Physical mass of C and S 𝑣𝑐 , 𝑣𝑠 : Instantaneous velocity of C and S
Visualisation of Risk Level
𝑘2 : Calibration coefficient
T: Collision prediction horizon
30 m/s, 1000 kg
30 m/s, 4000 kg
10 m/s, 1000 kg
30 m/s, 1000 kg Lane change to Left
𝑗: Road position of S at the end of prediction horizon, T 𝑃(𝐶𝑜𝑙𝑙𝑖𝑠𝑖𝑜𝑛
𝑇 𝑐,𝑗 ):
Probability of collision by C at j within the prediction horizon T.
𝜃𝑐,𝑗 : Angle between the direction of 𝑣𝑐 and 𝑟𝑐,𝑗 𝜃𝑠,𝑐 : Angle between the direction of 𝑣𝑐 and 𝑟𝑐,𝑗 𝑟𝑐,𝑗 : Vector from to initial road position of C to j
Vehicle Motion
• •
Point mass vehicle model The vehicle motion manipulated with independent lateral and longitudinal accelerations
•
The acceleration is constrained by the engine powertrain limitations, tyre force saturation limits
•
The vehicle velocity subject to non-holonomic constraint
Influence of neighbour vehicle C on the risk level of surrounding road space
FOR MIXED HIGHWAY TRAFFIC Application 1: Offline Safety Assessment
•
Offline Safety Assessment is the risk quantification approach of
Application 2: Online Risk Estimation Ex-ante trajectory risk estimation in highway merging scenario
realised trajectories assuming the knowledge of vehicle 15 m
capabilities
• •
12 m
Vehicle S may / may not perceive the risk posed by vehicle C
13 m/s
13 m/s
The risk perceived by S depends on its perception and
S
anticipation capability.
•
C2
C1
10 m/s
Ratio of risk measure to perceived risk measure by S is called
Merge scenario description at time t = 0 s
Perceived Risk Ratio
Merge Trajectory Highway cut-in situation
•
Subject vehicle S considers 3 merging trajectories T1, T2 & T3
Human driver: Alert and can anticipate a cut-in depending on
with different merge initiation time, but same acceleration profile
•
Acceleration in 𝑚/𝑠 2
the on/off status of turn indicator ACC equipped vehicle: Equipped with radar sensor (range & range rate sensing) and cannot predict a vehicle cut-in
•
Vehicle C: Human driven and right lane change indicator ON
1
Longitudinal acceleration
0.5 0 0 -0.5 -1
C
1
2
3
4
5
Lateral acceleration
Time from merge initiation in 𝑠
Prediction set up
• •
S
Table 1. Instantaneous risk measures by S (as human driven
Prediction time step: 0.1 s, prediction horizon: 11 s C1 & C2 maintain constant velocity
Risk of different trajectories
vehicle & ACC equipped vehicle) in a cut-in situation
Safety Indicator
Human (in Kg2)
ACC (in Kg2)
Time To Collision
NA
NA
Risk Measure
1.16 x 10-2
1.16 x 10-2
Perceived Risk Measure
1.16 x 10-2
3.32 x 10-7
Perceived Risk Ratio
1
2.8 x 10-5
T2: Merging begins at 1.5 s
T1: Merging begins at 1 s T3: Merging begins at 6 s, after C1 passing
Risk profiles for three alternate merge trajectories
Contribution A novel risk assessment approach for traffic including automated vehicles. The risk measure:
• • • • •
describes risk continuously in a 2-D vehicle interaction is defined specific to a vehicle type based on its manoeuvre variability and perception capability incorporates collision probability and severity represents the kinematic collision mechanism is capable of risk description of all near-miss situations even those created without a collision/ crossing course
This work is supported by NWO Domain TTW, the Netherlands, under the project “From Individual Automated Vehicles to Cooperative Traffic Management - Predicting the benefits of automated driving through on-road human behavior assessment and traffic flow models (IAVTRM)”- TTW#13712.
Freddy A. Mullakkal-Babu, Meng Wang, Bart van Arem, Riender Happee
Research Problem Surrogate safety measures are inadequate for safety assessment of mixed traffic fleet including automated vehicles:
• • •
Defined not specific to a vehicle type, considering its manoeuvre
Objective
variability
•
Limited consideration for collision severity
To specify an anticipative and probabilistic risk assessment approach that is applicable to on-road mixed vehicle interactions
•
Discontinuous in two-dimensional vehicle encounter
To derive a risk measure that can be objectively interpreted and compared
Risk Definition 𝑹𝑰𝑺𝑲 𝑴𝑬𝑨𝑺𝑼𝑹𝑬𝒔,𝒄 = 𝑷(𝑪𝒐𝒍𝒍𝒊𝒔𝒊𝒐𝒏
𝑻 𝒄,𝒋
⋅ 𝑴𝒄 ⋅ 𝒆
𝒌𝟐 ⋅|𝒗𝒄 (𝒕)|⋅𝐜𝐨𝐬 𝜽𝒄,𝒋
⋅ 𝑴𝒔 ⋅ 𝒆
Neighbour vehicle: C
−𝒌𝟐 ⋅|𝒗𝒔 (𝒕)|⋅𝐜𝐨𝐬 𝜽𝒔,𝒄
Subject vehicle: S cCollision probability
Collision severity
𝑥 𝑗 , 𝑦 𝑗 : future road position of S
𝑅𝐼𝑆𝐾 𝑀𝐸𝐴𝑆𝑈𝑅𝐸𝑠,𝑐 is the collision risk measured by the subject
Modelling Assumptions
vehicle S due to a neighbouring vehicle C. In a multi-vehicle
•
encounter, the risk measured by S is the sum of risk contributed by individual vehicles involved in the encounter.
Collision severity increases with instantaneous vehicle velocity and physical mass of interacting vehicles
•
The probability distributions of lateral and longitudinal acceleration of a vehicle type can be estimated
𝑀𝑐 , 𝑀𝑠 : Physical mass of C and S 𝑣𝑐 , 𝑣𝑠 : Instantaneous velocity of C and S
Visualisation of Risk Level
𝑘2 : Calibration coefficient
T: Collision prediction horizon
30 m/s, 1000 kg
30 m/s, 4000 kg
10 m/s, 1000 kg
30 m/s, 1000 kg Lane change to Left
𝑗: Road position of S at the end of prediction horizon, T 𝑃(𝐶𝑜𝑙𝑙𝑖𝑠𝑖𝑜𝑛
𝑇 𝑐,𝑗 ):
Probability of collision by C at j within the prediction horizon T.
𝜃𝑐,𝑗 : Angle between the direction of 𝑣𝑐 and 𝑟𝑐,𝑗 𝜃𝑠,𝑐 : Angle between the direction of 𝑣𝑐 and 𝑟𝑐,𝑗 𝑟𝑐,𝑗 : Vector from to initial road position of C to j
Vehicle Motion
• •
Point mass vehicle model The vehicle motion manipulated with independent lateral and longitudinal accelerations
•
The acceleration is constrained by the engine powertrain limitations, tyre force saturation limits
•
The vehicle velocity subject to non-holonomic constraint
Influence of neighbour vehicle C on the risk level of surrounding road space
FOR MIXED HIGHWAY TRAFFIC Application 1: Offline Safety Assessment
•
Offline Safety Assessment is the risk quantification approach of
Application 2: Online Risk Estimation Ex-ante trajectory risk estimation in highway merging scenario
realised trajectories assuming the knowledge of vehicle 15 m
capabilities
• •
12 m
Vehicle S may / may not perceive the risk posed by vehicle C
13 m/s
13 m/s
The risk perceived by S depends on its perception and
S
anticipation capability.
•
C2
C1
10 m/s
Ratio of risk measure to perceived risk measure by S is called
Merge scenario description at time t = 0 s
Perceived Risk Ratio
Merge Trajectory Highway cut-in situation
•
Subject vehicle S considers 3 merging trajectories T1, T2 & T3
Human driver: Alert and can anticipate a cut-in depending on
with different merge initiation time, but same acceleration profile
•
Acceleration in 𝑚/𝑠 2
the on/off status of turn indicator ACC equipped vehicle: Equipped with radar sensor (range & range rate sensing) and cannot predict a vehicle cut-in
•
Vehicle C: Human driven and right lane change indicator ON
1
Longitudinal acceleration
0.5 0 0 -0.5 -1
C
1
2
3
4
5
Lateral acceleration
Time from merge initiation in 𝑠
Prediction set up
• •
S
Table 1. Instantaneous risk measures by S (as human driven
Prediction time step: 0.1 s, prediction horizon: 11 s C1 & C2 maintain constant velocity
Risk of different trajectories
vehicle & ACC equipped vehicle) in a cut-in situation
Safety Indicator
Human (in Kg2)
ACC (in Kg2)
Time To Collision
NA
NA
Risk Measure
1.16 x 10-2
1.16 x 10-2
Perceived Risk Measure
1.16 x 10-2
3.32 x 10-7
Perceived Risk Ratio
1
2.8 x 10-5
T2: Merging begins at 1.5 s
T1: Merging begins at 1 s T3: Merging begins at 6 s, after C1 passing
Risk profiles for three alternate merge trajectories
Contribution A novel risk assessment approach for traffic including automated vehicles. The risk measure:
• • • • •
describes risk continuously in a 2-D vehicle interaction is defined specific to a vehicle type based on its manoeuvre variability and perception capability incorporates collision probability and severity represents the kinematic collision mechanism is capable of risk description of all near-miss situations even those created without a collision/ crossing course
This work is supported by NWO Domain TTW, the Netherlands, under the project “From Individual Automated Vehicles to Cooperative Traffic Management - Predicting the benefits of automated driving through on-road human behavior assessment and traffic flow models (IAVTRM)”- TTW#13712.
