Continual Humanoid Motion Learning

2026-10-03

AI Summary

This study addresses the problem of catastrophic forgetting when humanoid robots continuously acquire new skills by proposing a similarity-guided LoRA-PNN strategy. The method introduces a novel two-level motion similarity metric based on dynamic time warping and optimal transport, which guides knowledge reuse and capacity allocation within progressive neural networks. By integrating low-rank adaptation for lightweight parameter updates, it effectively prevents forgetting at the architectural level. Experimental results demonstrate that the proposed approach achieves state-of-the-art forward transfer performance while reducing parameter count by 94.5% and training time by 40.8%. Furthermore, it attains a Sim-to-Sim transfer rate of 96.13% and has been successfully deployed on a physical Unitree G1 robot.

Abstract

Humanoid whole-body controllers can now track a diverse set of dynamic motions, but they are typically trained offline and then frozen, so teaching such a controller a new skill tends to erode the skills it already mastered. We study continual learning for humanoid whole-body motion, where a single controller must acquire skills from a sequential task stream without revisiting past data. We introduce Similarity-guided LoRA-PNN, a progressive neural network (PNN) policy that prevents catastrophic forgetting by construction while reusing knowledge across skills through lightweight low-rank adaptation. A two-level motion-similarity measure, built from dynamic time warping aggregated by optimal transport, decides which prior skill to build on and how much new capacity to allocate, yielding strong forward transfer and large efficiency gains. Across six sequentially learned skill categories, our similarity-guided LoRA policy attains the best forward transfer (0.125 vs. 0.079) and the highest average accuracy among all methods, while saving up to 94.5% of trainable parameters and 40.8% of training time. The resulting controller reaches 96.13% sim-to-sim transfer and is deployed on a physical Unitree G1. Our code is available at https://anonymous.4open.science/r/continual-humanoid-learning-35D3.

Problem

• Continual Learning
• Humanoid Motion
• Catastrophic Forgetting
• Whole-body Control

Innovation

• Continual Learning
• Low-Rank Adaptation (LoRA)
• Progressive Neural Networks
• Dynamic Time Warping
• Humanoid Motion Control