Learning Autonomous and Safe Quadruped Traversal of Complex Terrains Using Multi-Layer Elevation Maps
Yeke Chen, Ji Ma, Zeren Luo, Yimin Han, YINZHAO DONG, Bowen Xu, Peng Lu
AI summary
Problem
Legged robots struggle to achieve robust, autonomous locomotion in cluttered and unstructured environments because existing terrain representations fail to capture confined spaces and real-world sensor data is often noisy or incomplete.
Approach
The authors introduce a hierarchical navigation system that processes a three-layer elevation map to represent complex terrains, using a trained neural compressor to convert noisy occupancy grids into this format, alongside terrain augmentation and knowledge distillation to build a robust locomotion policy.
Key results
- Novel three-layer elevation map effectively captures overhanging obstacles and confined spaces
- Neural terrain compressor accurately converts noisy occupancy grids to elevation maps with low error
- Hierarchical navigation achieves superior generalization and maneuverability in unseen randomized terrains
- Successful real-world deployment on a low-cost quadruped robot in indoor and outdoor environments
Why it matters
This framework provides a practical, perception-robust solution for autonomous legged navigation, advancing real-world deployment for inspection, exploration, and rescue applications.
Abstract
Legged robots hold great promise for agile and flexi- ble mobility across diverse and unstructured terrains, inspired by the remarkable adaptability of bipeds and quadrupeds in nature. However, achieving robust autonomous locomotion in cluttered and complex environments remains a significant challenge. In this work, we present a hierarchical control framework for quadrupedal robots that enables safe and autonomous traversal of cluttered terrains. Central to our approach is a novel multi- layer elevation map representation, which is generalized enough to capture a wide range of terrains. To further improve policy generalization and maneuverability, we incorporate terrain aug- mentation, knowledge distillation, and carefully designed reward functions. Extensive simulation experiments demonstrate that each component contributes to improved policy generalization, and that our terrain representation is more efficient and informa- tive than existing alternatives. By training a terrain compressor in simulation, we successfully deploy our system on a low-cost quadrupedal robot in real-world environments, showcasing the practicality and robustness of our approach.