A Multiresolution A* Method For Robot PathPlanning

摘要

In this paper, a point-to-point robot path planning problem is studied. It occurs in industry for example in spot welding, riveting, and pick and place tasks. A new A -based method is presented. The algorithm searches the robot's con guration space with many di erent resolutions at the same time. When a path candidate goes far from the obstacles, coarser resolutions corresponding to bigger step sizes is used. When it goes near the obstacle surfaces, ner resolutions corresponding to smaller step sizes is used. The algorithm always nds a path from a starting robot con guration to the goal con guration if one exists, which is a property of the A search in general. This is true given the nest resolution of the search space. These kind of path planning algorithms are called resolution complete in the robotics literature. A is also applied because of the possibility to generate better guiding heuristics. A better admissible heuristic roughly means that A using it expands fewer con guration space nodes, which is known a priori. A known AI-method utilizing relaxed models is applied to generate admissible heuristics. Constructing relaxed models involves removing details from the base level problem to get simpli ed ones. The heuristics are then obtained by solving these simpli ed problems. A simulated robot workcell is provided for demonstrations. The path planning of a 5-degrees-of-freedom industrial robot appears to be reasonably fast.