On-Board Perception and Motion Planning for Legged Locomotion over Rough Terrain.

Abstract

This paper addresses the issues of perception and motion planning in a legged robot walking over a rough terrain. The solution takes limited perceptual capabilities of the robot into account. A passive stereo camera and a 2D laser scanner are considered for sensing the terrain, with environment representations matching the uncertainty characteristics of these sensors. The motion planner adheres to the dual-representation concept by applying the A⋆ algorithm for strategic level planning on a coarse elevation grid from stereo data, and then using a RRTbased method to find a sequence of feasible motions on a more precise but smaller map obtained by using the laser data. Results of simulations and experiments on real data are provided.