Learning Roller-Skating Motions of Humanoid Robots Based on Adversarial Motion Priors

2026-07-12

AI Summary

Humanoid robots face significant challenges in roller skating, including whole-body balance, rolling contact dynamics, and velocity-dependent coordination of posture. This work proposes the first application of the Adversarial Motion Priors (AMP) reinforcement learning framework to this domain. Leveraging motion capture data—processed through retargeting, smoothing, and resampling—the method generates reference trajectories to train two complex skating gaits: Pump Glide and Push Glide, each guided by tailored reward functions. Experimental results demonstrate that the approach achieves high-quality gait generation, accurate velocity tracking, and agile turning in simulation. Ablation studies further validate the contribution of individual reward components, highlighting AMP’s potential for learning sophisticated dynamic motor skills in humanoid systems.

Abstract

Humanoid roller-skating is difficult because the robot must coordinate whole-body balance, rolling contacts, and velocity-dependent posture regulation. This paper presents an adversarial motion prior based reinforcement learning framework for two humanoid roller-skating gaits: Pump Glide skating and Push Glide skating. The two gait datasets are collected independently through motion capture and retargeted to the humanoid robot separately. The retargeted data are then smoothed and resampled into reference motion states for AMP training. The two gaits are learned by independent AMP training pipelines with separate reference datasets, separate policies, and independent reward architectures. Simulation experiments are designed to evaluate gait quality, velocity tracking, turning, and gait-specific reward ablations.

Problem

• humanoid roller-skating
• whole-body balance
• rolling contacts
• velocity-dependent posture regulation
• motion learning

Innovation

• Adversarial Motion Priors
• Humanoid Roller-Skating
• Reinforcement Learning
• Motion Retargeting
• Gait Learning