220 likes | 240 Views
Optimal Planning for Vehicles with Bounded Curvature: Coordinated Vehicles and Obstacle Avoidance. Andy Perrin. Two Problems. Optimal Planning for multiple vehicles. Paper by Antonio Bicci and Lucia Pallottino Optimal Planning for obstacle avoidance.
E N D
Optimal Planning for Vehicles with Bounded Curvature:Coordinated Vehiclesand Obstacle Avoidance Andy Perrin
Two Problems • Optimal Planning for multiple vehicles. • Paper by Antonio Bicci and Lucia Pallottino • Optimal Planning for obstacle avoidance. • My variation (so to speak) on the above paper.
Dubin’s Problem • Given an initial position and direction, and a final position and direction, find the shortest path from one to the other, given that the curvature is bounded.
Dubins’ Solution • All Dubins’ paths consist of circular arcs and straight lines. • Arcs will have the minimum curvature (which is specified in the problem statement) • For least time problems, speed = max speed
Multiple Vehicles • Consider a Dubins’-type problem with n vehicles, and require that they not collide. • If we wish to minimize the overall time, what paths should they take, and how should we schedule them?
Problem Statement R1 1 d13 3 d12 R3 d23 2 R2
Problem Restatement • Can re-write this as an example of Pontryagin’s Minimization Principle. (PMP) • Authors do this, find a Hamiltonian, take variations, and find necessary conditions for stationary solutions. • Solutions consist of unconstrained arcs, and constrained arcs.
Case of no collisions (ie unconstrained arcs only) • Simply calculate the Dubins’ path for each vehicle independently.
Case of some collisions • Authors decompose the possible types of collisions into several categories of constrained arcs, and find the solution for each category. • The arcs are then stitched together with some matching conditions. • Dubins’ paths are extremely fast to calculate!
Advantages + Disadvantages • Advantages: • Very Fast • Global optimum found • Smooth paths • Disadvantages: • No obvious way to include obstacles • # of sub-problems increases combinatorially with # of vehicles • Vehicles can’t go backwards (but others have worked on this) • Feed-forward
Optimal Planning with Obstacles • I concentrated on modeling a single vehicle with arbitrary obstacles. • But, my method extends naturally to the n-vehicle case.
Parametize then Optimize! • Given a starting and an ending point, I divide the path into n-1 pieces. • Velocity is assumed constant on each piece. • I add up the total time T, and minimize with fmincon, subject to some constraints…
How to Represent Obstacles? • Obstacles are simply contour surfaces (“blobbies” or “metaballs”), analogous to the sliding surfaces we talked about earlier in the semester. • Typically, sums of Gaussian distributions are used, but this is not necessary.
And a Small Problem… Local Minima!
Problems and Future Improvements • Problems • Number of design variables increases quickly • Local minima • Speed is slow compared to Bicci paper. • Possible Improvements • Change from using line segments to cubic splines (Hermite polynomials) • Multiple vehicles