Computing Reachable Sets via Toolbox of Level Set Methods Michael Vitus (michael.vitus@gmail.com) Jerry Ding (jding@eecs.berkeley.edu) 4/16/2012 Toolbox of Level Set Methods • Ian Mitchell – Professor at the University of British Columbia • http://www.cs.ubc.ca/~mitchell/ • Toolbox – Matlab – Computes the backwards reachable set – Fixed spacing Cartesian grid – Arbitrary dimension (computationally limited) Problem Formulation • Dynamics: – System input: – Disturbance input: • Target set: – Unsafe final conditions Backwards Reachable Set • Solution to a Hamilton-Jacobi PDE: where: • Terminal value HJ PDE – Converted to an initial value PDE by multiplying the H(x,p) by -1 Toolbox Formulation • No automated method • Provide 3 items – Hamiltonian function (multiplied by -1) – An upper bound on the partials function – Final target set General Comments • Hamiltonian overestimated reachable set underestimated • Partials function – Most difficult – Underestimation numerical instability – Overestimation rounded corners or worst case underestimation of reachable set • Computation – The solver grids the state space – Tractable only up to 6 continuous states • Toolbox – Coding: ~90% is setting up the environment Useful Dynamical Form • Nonlinear system, linear input • Input constraints are hyperrectangles • Analytical optimal inputs: • Partials upper bound: Example: Two Identical Vehicles b • Kinematic Model • Target set – Protected zone y xr – Position and heading angle – Inputs: turning rates v r yr Blunderer v r a Evader Example • Optimal Hamiltonian: • Partials: Results Toolbox • Plotting utilities – Kernel\Helper\Visualization – visualizeLevelSet.m – spinAnimation.m • Initial condition helpers – Cylinders, hyperrectangles • Advice: Start small… • Walk through example