Decentralized Motor Skill Learning for Complex Robotic Systems
This work addresses robustness and generalization issues in robotic control for applications like locomotion, though it is incremental as it builds on existing RL methods with a decentralized twist.
The paper tackles the problem of centralized control in robotic reinforcement learning, which leads to task-specific policies vulnerable to local disturbances, by proposing a decentralized motor skill learning algorithm that automatically discovers decoupled motor groups and learns a decentralized policy, resulting in improved robustness and generalization without performance loss, as demonstrated on quadruped and humanoid robots.
Reinforcement learning (RL) has achieved remarkable success in complex robotic systems (eg. quadruped locomotion). In previous works, the RL-based controller was typically implemented as a single neural network with concatenated observation input. However, the corresponding learned policy is highly task-specific. Since all motors are controlled in a centralized way, out-of-distribution local observations can impact global motors through the single coupled neural network policy. In contrast, animals and humans can control their limbs separately. Inspired by this biological phenomenon, we propose a Decentralized motor skill (DEMOS) learning algorithm to automatically discover motor groups that can be decoupled from each other while preserving essential connections and then learn a decentralized motor control policy. Our method improves the robustness and generalization of the policy without sacrificing performance. Experiments on quadruped and humanoid robots demonstrate that the learned policy is robust against local motor malfunctions and can be transferred to new tasks.