Exploiting collision information in probabilistic roadmap planning

This paper develops a novel approach to combining probabilistic motion planners. Rather than trying to develop a single planner that works over a wide range of environments, we develop a strategy for combining different motion planners within a single framework. Specifically we examine how planners designed for open spaces and those designed for narrow passages can be integrated within a single planning framework. Information that is normally discarded in the planning process is used to identify regions as being potentially 'narrow' or 'cluttered', and we then apply the planner most suited for that region based on this information. Experimental results demonstrate our approach outperforms the basic PRM approach as well as a Gaussian sampler designed for narrow regions in three test environments.