DAMP: Humanoid Locomotion via Denoised Belief Learning and Adversarial Motion Priors

2026-10-08

AI Summary

This study addresses the challenge of achieving stable locomotion for humanoid robots in complex terrains without external perception. To this end, it proposes DAMP, a reinforcement learning framework that innovatively integrates denoising belief learning with adversarial motion priors. By leveraging recurrent neural networks to capture temporal dependencies and infer latent privileged information, the method enables implicit state estimation and policy alignment, thereby achieving robust and natural end-to-end motion control. Furthermore, through sim-to-real transfer techniques, the proposed approach successfully facilitates zero-shot deployment from simulation to the real world, significantly enhancing both the robustness and generalization capability of robotic locomotion.

Abstract

Humanoid robots possess the structural capability to traverse complex terrains. However, achieving stable t raversal without relying on perceived information remains challenging, particularly in complex environments. This paper introduces DAMP, a reinforcement learning framework aimed at achieving robust and naturalistic humanoid locomotion over challenging terrains, with the assumption that no perceived information is available. The framework leverages recurrent neural networks to capture temporal dependencies and implicitly infer privileged and other task-relevant latent information. By aligning the learned representations with the task objective, the method enables robust and goal-consistent policy learning. This end-to-end framework achieves transfer learning from simulation to real-world environments, demonstrating the proposed method's robustness and generalization capabilities. The video of the real-world demonstration can be found at the following link: https://youtu.be/AkI7TZB2DDM.

Problem

• Humanoid Locomotion
• Complex Terrains
• Blind Navigation
• Reinforcement Learning

Innovation

• Humanoid Locomotion
• Denoised Belief Learning
• Adversarial Motion Priors
• Recurrent Neural Networks
• Sim-to-Real Transfer