Using manipulability to bias sampling during the construction of probabilistic roadmaps

Abstract

Probabilistic roadmaps (PRMs) are a popular representation used by many current path planners. Construction of a PRM requires the ability to generate a set of random samples from the robot's configuration space, and much recent research has concentrated on new methods to do this. In this paper, we present a sampling scheme that is based on the manipulability measure associated with a robot arm. Intuitively, manipulability characterizes the arm's freedom of motion for a given configuration. Thus, our approach is to densely sample those regions of the configuration space in which manipulability is low (and therefore, the robot has less dexterity), while sampling more sparsely those regions in which the manipulability is high. We have implemented our approach, and performed extensive evaluations using prototypical problems from the path planning literature. Our results show this new sampling scheme to be effective in generating PRMs that can solve a large range of path planning problems.