Neuromechanics-Based Reinforcement Learning for FES Control of Lower-Limb Movements

Stroke often leads to motor dysfunction and gait impairment, significantly affecting patients' mobility and quality of life. Functional electrical stimulation (FES) has been widely used in neurorehabilitation to activate paralyzed muscles and promote motor recovery. However, achieving coordinated control of multiple lower-limb joints remains challenging due to complex musculoskeletal dynamics. This study proposes a control framework that integrates a musculoskeletal model with reinforcement learning to generate muscle stimulation signals for coordinated lower-limb movements. The musculoskeletal model simulates the dynamics of the hip, knee, and ankle joints, while reinforcement learning is used to learn optimal stimulation strategies for joint control. Simulation results show that the proposed method can achieve coordinated control of the three joints under FES actuation. The root mean square errors (RMSE) of the hip, knee, and ankle joints were 3.52°, 4.60°, and 1.72°, respectively. Compared with a conventional controller, the reinforcement learning-based method improved trajectory tracking accuracy and produced smoother joint movements, demonstrating its potential for FES-based lowerlimb rehabilitation.