ADP: Adversarial Dynamics Priors for Physically Grounded Humanoid Locomotion

2026-07-03

AI Summary

Existing motion-prior-based methods model only kinematic features while neglecting dynamic characteristics—such as center-of-mass dynamics, momentum, and contact forces—resulting in insufficient disturbance rejection for humanoid robots. This work proposes Adversarial Dynamics Priors (ADP), which extends adversarial imitation learning from the kinematic to the dynamic domain for the first time. ADP constructs a reference distribution centered on centroidal dynamics through trajectory optimization and employs a temporal window discriminator to guide policy learning within a reinforcement learning framework that does not require explicit action tracking, thereby generating dynamically consistent and robust motions. Compared to the strongest baseline, AMP, ADP improves the impact threshold at 80% success rate by 16.7%, reduces average recovery time by 47.9%, and decreases velocity tracking error by 35.4%.

Abstract

In this paper, we propose Adversarial Dynamics Priors (ADP) for perturbation-resilient humanoid locomotion control. Existing motion prior-based methods induce natural motion styles by imitating kinematic motion features, but they do not directly regularize dynamics features, such as CoM motion, centroidal momentum, contact forces, and contact states. To address this limitation, we replace kinematic motion-style feature with selected dynamics features extracted from locomotion trajectories as the target of adversarial regularization.To this end, we use trajectory optimization to construct a reference dataset and train a discriminator to evaluate whether policy-induced temporal windows are consistent with the resulting reference distribution.Without explicit motion tracking, ADP encourages policy rollouts to remain close to the reference support, even after perturbations. Experimental results show that, compared with AMP, the strongest baseline in our evaluation, ADP improves the $80\%$-success impulse threshold ($J_{80}$) by $16.7\%$, while reducing direction-averaged recovery time and velocity tracking error by $47.9\%$ and $35.4\%$, respectively.

Problem

• humanoid locomotion
• motion priors
• dynamics regularization
• perturbation resilience
• adversarial learning

Innovation

• Adversarial Dynamics Priors
• humanoid locomotion
• dynamics regularization
• perturbation resilience
• trajectory optimization