Higher-Order Morphology Priors for Quadruped Reinforcement Learning Under Actuator Degradation

2026-10-07

AI Summary

This study addresses the challenges of impaired locomotion coordination and insufficient compensation in quadruped robots caused by actuator degradation, proposing a whole-body control method based on higher-order topological modeling. The approach introduces cell complexes to represent second-order structural units of limbs and torso, incorporating morphological priors as inductive biases into the model. Furthermore, it designs a node-edge-face Actor policy network leveraging Hodge message passing to enable efficient multi-scale topological feature aggregation and policy generation. Experimental results demonstrate that the proposed method achieves optimal rewards in unseen degradation scenarios, significantly improving survival rates while reducing velocity tracking errors. These findings validate its strong generalization and compensation capabilities under actuator failures.

Abstract

Actuator degradation turns quadruped locomotion into a coordination problem requiring joints to compensate for lost actuation. Prior work suggests that morphology-aware graph policies improve learning and generalization under body perturbations. We ask whether these benefits can be strengthened by explicitly modeling higher-order mechanical structure. We represent the Unitree Go1 as a cell complex with limb- and body-level rank-2 cells and apply Hodge-based message passing. Under degradation training, the node-edge-face Hodge actor achieves the highest return on unseen actuator degradations, with higher survival and lower velocity-tracking error. These results support higher-order morphology as a useful inductive bias for whole-body compensation under actuator degradation.

Problem

• Quadruped locomotion
• Actuator degradation
• Higher-order morphology
• Reinforcement learning

Innovation

• Cell Complex
• Hodge Theory
• Actuator Degradation
• Morphology Priors
• Quadruped Reinforcement Learning