Workspace-based sampling for probabilistic path planning

Abstract

This thesis focuses on probabilistic path planning, the most promising approach for path planning of robots with many degrees of freedom. Despite its success, probabilistic path planners perform poorly when the configuration space contains narrow passages. Our thesis is workspace provides useful information for generating suitable sampling distributions for probabilistic path planning. We start by presenting an empirical study to explore the importance of sampling distribution in probabilistic path planning. We then explore the relation between workspace and configuration space of the robot. Finally, we present our new workspace-based probabilistic path planner Workspace-based Connectivity Oracle (WCO). Analysis shows that WCO is probabilistic complete. Experimental results show that WCO significantly improves the performance of probabilistic path planners in solving realistic narrow passages problems. The applicability of WCO is also shown in a simulation of the bridge-inspection problem involving a hyper-redundant robot with up to 35 DOFs.