Adaptive Hybrid Sampling for Probabilistic Roadmap Planning

Several sophisticated sampling strategies have been proposed recently to address the narrow passage problem for probabilistic roadmap (PRM) planning. They all have unique strengths and weaknesses in different environments, but none seems sufficient on its own in general. In this paper, we propose a systematic approach for adaptively combining multiple sampling strategies for PRM planning. Using this approach, we describe three adaptive hybrid sampling strategies. Two are motivated by theoretical results from the computational learning theory. Another one is simple and performs well in practice. We tested them on robots with two to eight degrees of freedom in planar workspaces. In these preliminary tests, the adaptive hybrid sampling strategies showed consistently good performance, compared with fixed-weight hybrid sampling strategies.