David Hsu, Robert Kindel, Jean- Claude Latombe, Stephen Rock Presented by: Haomiao Huang Vijay Pradeep Randomized Kinodynamic Motion Planning with Moving Obstacles
Planner Overview - Account for robot’s kinematics & dynamics - Use a forward dynamics model - Plan in state x time space -Avoid moving obstacles Robot Obstacle Planner Overview
Straight Line Segments 1) Randomly Generate New Milestone 2) Try to connect the milestone to existing milestones Traditional PRM
Planner Overview 1) Choose an existing milestone 2) Generate a new milestone using a random control input t=1 t=2 t=1 t=3 t=2 Goal Region Current State Terminate When Milestone reaches Goal Region Control Sampling PRM
Planner Overview Goal Region Sampling Strategy Weighted sampling approximates ideal sampling
Planner Overview Goal Region Terminate When Milestone in first tree is “close enough” to milestone in second tree Forward & Backward Integration
Planner Overview Probabilistic Completeness & Exponential Convergence - Volume of reachable set exponentially bounded by number of lookout points - Probability of lookout points increases exponentially with number of milestones - Probability of finding a solution increases exponentially with number of milestones Expansiveness: Visibility Becomes Reachability
Planner Overview Car Like Robots
Planner Overview Running Times Histogram Run Times, Collision Checks, Milestones, And Propagations Car Like Robots
Planner Overview Double Integrator Dynamics, Moving Obstacles x y ARL Air-Cusion Robot
Planner Overview x y t Moving Obstacles
Planner Overview “Real-Time” Planning – Escape Trajectories Goal Safe Region Not always possible to find solution to goal in allotted computation time Robot
Planner Overview “Real Time” Planning – Time delays Computing the trajectory also takes time Robot Δt plan =0.4 sec Instantaneous planning Propagating dynamics by Δt plan
Planner Overview Actual Obstacle Trajectory “Real Time” Planning – On-the-fly Replanning Robot Estimated Obstacle Trajectory Planned Trajectory
Planner Overview Path Planning with Moving Obstacles
Planner Overview Conclusions Kinodynamic constraints can be dealt with through input sampling Expansiveness can be generalized to kinodynamic configuration spaces through reachability Moving obstacles can be efficiently dealt with “Real-Time” Planning is tricky to do well Issues: –Narrow Passages? –Long tail of running time