A Random Path Sampling-Based Method for Motion Planning in Many Dimensions

Abstract

In this paper, we present GRPMP (Gaussian Random Paths Motion Planner), a method for solving robot motion planning problems. Compared with traditional sampling-based methods which build large-scale random trees or roadmaps by sampling random nodes in configuration-spaces, GRPMP generates a set of smooth random paths discretized into sparse nodes, using a random path sampler based on Gaussian process, to reduce resource space occupation and improve planning success rate. Employing the high-quality path selecting strategy with the lazy collision checking technique, GRPMP can fast select the collision-free high-quality path that has the lowest cost in the set of smooth random paths, so as to improve the efficiency and quality of the motion planning. We present challenging experiments with GRPMP on pick-and-place tasks, using a humanoid robot, to show that GRPMP has the ability to find high-quality paths with efficiency and stability.