Adaptive Strategies for Probabilistic Roadmap Construction

This paper presents an experimental study of prospects for using adaptable local search techniques in probabilistic roadmap based motion planning. The classical PRM approach uses a single fast and simple local planner to build a network representation of the configuration space. Advanced PRM planners utilize heuristic sampling techniques and combine multiple local planners. The planner described here uses a single local planner, but adjusts its competence during the roadmap construction stage according to the problem at hand. Two adjusting strategies are proposed and compared experimentally against using a static local planner at a set competence level. The results indicate that roadmap construction with an adaptive local planner can bring advantages including more robust performance and a reduction in planning cost variance.