An obstacle-based probabilistic roadmap method for path planning

Abstract

This thesis presents a new obstacle-based probabilistic roadmap method for motion planning for many degree of freedom robots that can be used to obtain high quality roadmaps even when the robot's configuration space is crowded. The main novelty in this approach is that roadmap candidate points are chosen on the constraint surfaces corresponding to obstacles in the workspace. As a consequence, the roadmap is likely to contain difficult paths, such as those traversing long, narrow passages in the robot's configuration space. The approach can be used for both collisionfree path planning and for planning of contact tasks. A path planner based on this approach is implemented for planar articulated robots in a two-dimensional workspace with polygonal obstacles. Experimental results with various types of robots and a range of environments show that well connected roadmaps are generated for most environments, even when the number of nodes in the roadmap is small. After the roadmap is built during preprocessing, many difficult path planning operations are carried out in less than a second.