Probabilistic roadmaps-putting it all together

Given a robot and a workspace, probabilistic roadmap planners (PRMs) build a roadmap of paths sampled from the workspace. A roadmap node is a single collision-free robot configuration, randomly generated. A roadmap edge is a sequence of collision-free robot configurations which interpolate the path from one roadmap node to another. Queries to the roadmap are (start, goal) pairs. If both the start and goal of a pair can be connected to the same connected component of the roadmap, the query is solved. Many promising variants of the PRM have been proposed, each with their own strengths and weaknesses. We propose a meta-planner for using many PRMs in such a way that the strengths are combined and the weaknesses offset. Our meta-planner will perform the combination in the following manner: i) provide a framework in which different motion planners are available and to which new ones are easily added; ii) characterize subregions (possibly overlapping) based on sample characteristics and connection results; iii) assign subregions to one or more planners which are judged promising; and iv) provide stopping criteria for roadmap construction. We present experimental results for four characterization measures. A general technique we call ´filtering´ is presented for keeping roadmaps compact.