Probabilistic path planning with extended local planners

Abstract

We present a path planning algorithm for a six-degree-of-freedom (6 DOF) polyhedral robot moving in a known and static environment. The planner is the dual-tree rapidly exploring random tree (RRT) algorithm that uses a novel local planner. A local planner is a subroutine that determines whether two robot configurations can be connected by a simple path. The local planner we develop searches a 2 DOF or a 3 DOF subspace of the configuration space, whereas prior planners search the line segment that connects the two configurations. Although our planner has a higher cost at the local planning level, the empirical data shows that our planner outperforms prior planners on problems with narrow channels and performs comparably on other problems.