Hardware-in-the-Loop for Robot Joint Module and Reinforcement Learning: A Case Study on Three-Phase Permanent Magnet Synchronous Motor

The paper shows a hardware-in-the-loop (HIL) framework for training reinforcement learning (RL) policies to control a three-phase permanent magnet synchronous motor (PMSM) in robotic joint modules. By integrating real-time motor dynamics with RL-driven control strategies, the paper address the challenges of dynamic adaptability in high-precision robotic systems under RL framework. The proposed RL policy leverages sampled currents, rotor position, and angular velocity to optimize torque output while regulating the current performance (i.e., energy consumption). Experimental results demonstrate a 22% improvement in trajectory tracking accuracy and a 15% reduction in torque ripple compared to traditional PD control.