Sensor-based probabilistic roadmaps: experiments with an eye-in-hand system

Abstract

We present a real implemented eye-in-hand test-bed system for sensor-based collision-free motion planning for articulated robot arms. The system consists of a PUMA 560 with a triangulation-based area-scan laser range finder (the eye) mounted on its wrist. The framework for our planning approach is inspired by recent motion planning research for the classical model-based case (known environment) and incrementally builds a roadmap that represents the connectivity of the free configuration space, as the robot senses the physical environment. We present some experimental results with our sensor-based planner running on this real test-bed. The robot is started in completely unknown and cluttered environments. Typically, the planner is able to reach (planning as it senses) the goal configuration in about 7-25 scans (depending on the scene complexity), while avoiding collisions with the obstacles throughout.