This paper proposes a novel framework that extends slope-heading energy models to enable multi-terrain gait adaptation for legged robots, integrating terrain classification, physics-informed energy prediction, and hierarchical reinforcement learning to dynamically select energy-optimal gaits.
Key findings
Proposes a Multi-Terrain Energy Model (MTEM) predicting energy consumption based on terrain type, slope angle, and heading direction.
Develops a Hierarchical Gait Adaptation Policy (HGAP) that learns to modulate gait parameters for energy-optimal locomotion.
Demonstrates potential energy savings of 25–40% compared to fixed-gait controllers while maintaining traversal success rates above 95%.
Limitations & open questions
The model's applicability to real-world terrains requires extensive experimental validation.