Randomized sampling with fixed and dynamic space decomposition methods

In this paper we introduce an adaptive sampling strategy for some of the more advanced probabilistic roadmap planners recently proposed. Each of these planners uses a distinct scheme to judiciously generate only well-positioned milestones that can improve the quality of the roadmap being constructed. However, the cost of generating each milestone for such a planner is usually very high. Our adaptive sampling strategy, using a fixed or dynamic decomposition of the robot's configuration space C, can guide these planners by identifying promising areas of C where (useful) milestones can be efficiently generated. Experimental results show that, for the two planners tested, the adaptive sampling strategy shortens the roadmap construction time by significantly reducing the milestone-generation cost.