Perceptive Locomotion in Rough Terrain – Online Foothold Optimization

摘要

Compared to wheeled vehicles, legged systems have a vast potential to traverse challenging terrain. To exploit the full potential, it is crucial to tightly integrate terrain perception for foothold planning. We present a hierarchical locomotion planner together with a foothold optimizer that finds locally optimal footholds within an elevation map. The map is generated in real-time from on-board depth sensors. We further propose a terrain-aware contact schedule to deal with actuator velocity limits. We validate the combined locomotion pipeline on our quadrupedal robot ANYmal with a variety of simulated and real-world experiments. We show that our method can cope with stairs and obstacles of heights up to 33% of the robot's leg length.