Humanoid robots learn new motions without forgetting old skills
Continual Humanoid Motion Learning
Robotics
Summary
Robots that move like humans often struggle to learn new tricks without forgetting the old ones. The authors developed a method that helps a robot keep all its learned movements while adding new ones one by one. They use a smart way to figure out which old skill is most like the new one and adapt only a small part of their system to learn the new motion. This approach saves time, reduces how much of the robot's brain needs to change, and works well both in simulations and on a real robot.
What this means in practice
- •For robotics engineers: Create humanoid robot controllers that can continuously learn new motions without losing previous skills, improving adaptability in dynamic environments.
- •For industrial robot operators: Update humanoid robots with new movement abilities efficiently, reducing downtime and retraining costs on factory floors.
Authors
Zhewen He, Hao Huang, Geeta Chandra Raju Bethala, Chong Yu, Tao Chen, Anthony Tzes, Yi Fang
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.