Learning Locomotion on Complex Terrain for Quadrupedal Robots with Foot Position Maps and Stability Rewards

2026-04-03

AI Summary

This work addresses the challenges of inaccurate foot placement and motion instability commonly encountered by quadrupedal robots navigating complex terrains. The authors propose a reinforcement learning approach that integrates explicit foothold position maps with elevation maps, incorporating—for the first time—a dynamic motion-stability reward mechanism within an attention-based framework to jointly optimize gait accuracy and overall stability. This method significantly improves locomotion success rates both within and beyond the training domain, enabling more precise and robust foothold control. By effectively bridging the gap in existing reinforcement learning strategies that often sacrifice either precision or stability, the proposed approach demonstrates superior performance in handling diverse and challenging terrain conditions.

Abstract

Quadrupedal locomotion over complex terrain has been a long-standing research topic in robotics. While recent reinforcement learning-based locomotion methods improve generalizability and foot-placement precision, they rely on implicit inference of foot positions from joint angles, lacking the explicit precision and stability guarantees of optimization-based approaches. To address this, we introduce a foot position map integrated into the heightmap, and a dynamic locomotion-stability reward within an attention-based framework to achieve locomotion on complex terrain. We validate our method extensively on terrains seen during training as well as out-of-domain (OOD) terrains. Our results demonstrate that the proposed method enables precise and stable movement, resulting in improved locomotion success rates on both in-domain and OOD terrains.

Problem

• quadrupedal locomotion
• complex terrain
• foot position
• stability
• reinforcement learning

Innovation

• foot position map
• stability reward
• quadrupedal locomotion
• complex terrain
• reinforcement learning