An Adaptive Control Method for Humanoid Robots Based on Fully-Actuated Systems

This paper addresses the motion control problem of humanoid robots operating in real-world scenarios frequently subjected to various complex disturbances. To tackle this challenge, we propose a reinforcement learning-based method for the automatic tuning of trajectory tracking controller parameters. While trajectory tracking controllers can be designed based on dynamic models using fully-actuated system theory, fixed-parameter controllers often exhibit limited robustness and adaptability in the presence of unknown and dynamic disturbances encountered in practical applications. The innovation of this work lies in integrating the strengths of reinforcement learning and model-based control to achieve automated parameter optimization for enhanced disturbance rejection. Simulation results demonstrate effective tracking of humanoid joint despite random position resets.