An Investigation Into the Application of an RL-GA-Based Multi-Modal Motion Somatosensory Optimization Control Strategy for a Novel Rehabilitation Robot

Abstract

The increasing prevalence of balance disorders presents significant challenges to rehabilitation therapy, prompting the development of rehabilitation robots as effective tools for motion training. To enhance the patient experience and rehabilitation outcomes in human-robot collaborative training, a motion control strategy based on reinforcement learning and genetic algorithms (RL-GA) with somatosensory enhancement is proposed and implemented in a novel rehabilitation robot. This study introduces a mathematical model of the human sensory system to quantify somatosensory feedback related to motion and integrates the washout algorithm (WA) into the robot's control system, facilitating the reproduction of somatosensation and motion simulation. Three typical rehabilitation modes—level walking, stair climbing, and stair descending—are selected, with natural gait features extracted as predefined trajectories. The WA parameters for each mode are optimized using RL-GA in various rehabilitation scenarios. Simulation results demonstrate that the washout filtering optimization method using RL-GA reduces theoretical somatosensory error by approximately 10% to 20% across all three rehabilitation modes, compared to traditional GA optimization. The experimental results further confirm the reliability and feasibility of the proposed method. The proposed approach enhances the realism of robot-assisted motion, thereby theoretically improving training effectiveness and accelerating the rehabilitation process.