Autonomous Driving and the ROAD System48.pdfB 1 B 2 B 3 B 5 B 4 The transition model Case 1: A...
1
m Probability Robustly-Safe Automated Driving (ROAD) Systems Changliu Liu, Graduate Student Researcher; Masayoshi Tomizuka, Professor Autonomous Driving and the ROAD System Layer 2: Learning and Decision Making The ROAD System Case Study Efficiency: Navigate to the destination in minimum time • Predict the future course for each surrounding vehicle (learning and prediction); • Find a trajectory in the safe region (decision making). Safety: Interact with surrounding vehicles properly Behavior 1 (B1) : Lane Following Behavior 2 (B2) : Lane Changing to the Left Behavior 3 (B3) : Lane Changing to the Right Behavior 4 (B4) : Lane Merging Behavior 5 (B5) : Lane Exiting Steady State Behavior Maneuvers Exiting Behavior The Optimization Problem in Decision Making 1. Solve the optimal control problem without the safety constraint 2. Check if the resulting trajectory violates the safety constraint 3. If no, execute the resulting trajectory 4. If yes, modify the trajectory to make it safe Baseline Planner Safety Planner Knowledge Kinematic Model Planned Trajectory + + Strategy Vehicle State Estimator Learning Center and Predictor Procedures in solving the optimization problem: To be learned To be designed Simplified The Multi-Agent System Learning and Prediction The probability of lane changing rises since the lateral speed is nonzero. After observing lane changing, the probability of lane following rises. B 1 B 2 B 3 B 5 B 4 The transition model Case 3: Overtaken by a fast vehicle Case 2: A slow front vehicle Case 1: A stationary obstacle Objective: speed tracking and lane following, e.g. the cost function penalizes 1) the deviation from the desired speed and 2) the deviation from the lane center (the target lane center is subject to change according to different strategies) Safety constraint: the distance d between the automated vehicle and the front and rear vehicles on the same lane should be greater than d min =40m. The safety index . φ = d 2 min - d 2 - ˙ d minimize a cost function for the automated vehicle subject to constraint on control, vehicle dynamics safety constraint regarding surrounding vehicles
Transcript of Autonomous Driving and the ROAD System48.pdfB 1 B 2 B 3 B 5 B 4 The transition model Case 1: A...
mProb
abili
ty
Robustly-Safe Automated Driving (ROAD) SystemsChangliu Liu, Graduate Student Researcher; Masayoshi Tomizuka, Professor
Autonomous Driving and the ROAD System
Layer 2: Learning and Decision Making
The ROAD System
Case Study
Efficiency: Navigate to the destination in minimum time
• Predict the future course for each surrounding vehicle (learning and prediction); • Find a trajectory in the safe region (decision making).
Safety: Interact with surrounding vehicles properly
Behavior 1 (B1) : Lane FollowingBehavior 2 (B2) : Lane Changing to the LeftBehavior 3 (B3) : Lane Changing to the RightBehavior 4 (B4) : Lane MergingBehavior 5 (B5) : Lane Exiting
Steady State Behavior
Maneuvers
Exiting Behavior
The Optimization Problem in Decision Making
1. Solve the optimal control problem without the safety constraint 2. Check if the resulting trajectory violates the safety constraint 3. If no, execute the resulting trajectory 4. If yes, modify the trajectory to make it safe
Baseline Planner
Safety PlannerKnowledge
Kinematic Model Planned
Trajectory++
Strategy
Vehicle State Estimator
Learning Center and Predictor
Procedures in solving the optimization problem:
To be learned
To be designed
Simplified
The Multi-Agent System Learning and Prediction
The probability of lane changing rises since the lateral speed is nonzero.
After observing lane changing, the probability of lane following rises.
B1
B2
B3
B5
B4
The transition model
Case 3: Overtaken by a fast vehicleCase 2: A slow front vehicleCase 1: A stationary obstacle
Objective: speed tracking and lane following, e.g. the cost function penalizes 1) the deviation from the desired speed and 2) the deviation from the lane center (the target lane center is subject to change according to different strategies) Safety constraint: the distance d between the automated vehicle and the front and rear vehicles on the same lane should be greater than dmin=40m. The safety index .� = d2min � d2 � d
minimize a cost function for the automated vehicle subject to constraint on control, vehicle dynamics
safety constraint regarding surrounding vehicles
