Using workspace information as a guide to non-uniform sampling in probabilistic roadmap planners

The probabilistic roadmap (PRM) planner is a popular method for robot motion planning problems with many degrees of freedom. How-ever, it has been shown that the method performs less well in situa-tions where the robot has to pass through a narrow passage in the scene. This is mainly due to the uniformity of the sampling used in the planner; it places many samples in large open regions and too few in tight passages. Cell decomposition methods do not have this disad-vantage, but are only applicable in low-dimensional configuration spaces. In this paper, a hybrid technique is presented that combines the strengths of both methods. It is based on a robot independent cell decomposition of the free workspace guiding the probabilistic sampling more toward the interesting regions in the configuration space. The regions of interest (narrow passages) are identified in the cell decomposition using a method we call watershed labeling. It is shown that this leads to improved performance on difficult planning problems in two- and three-dimensional workspaces. KEY WORDS—path planning, probabilistic roadmaps, narrow-passage problem 1