A Robust Control Strategy Based on Reinforcement Learning Approach to Rehabilitat theArm Movement

In this research، a control strategy has been presented to movement control of a three link model of human’s arm. The freezing mechanism has been used to consider the role of antagonistic coactivation of wrist muscles in the used three link model. Inspired by motor learning process of central nervous system، the presented control strategy has been designed based on the reinforcement learning algorithm. At first، the performance of a control methodology based on reinforcement learning was evaluated. The results show the instability of control system even after numbers of leaning episode. Then، a combination of a proportional derivative integral (PID) controller and a reinforcement learning based controller were utilizedtoimprove the stability conditions and performance of controller. Despite the good performance، there is no guarantee for stability of control system. So، to satisfy the stability conditions، a robust controller called HTC was added to thecombination of a PID controller and a reinforcement learning based controller. According to the simulation results، the combinational controller accompany by HTC had good performance even in presence of external disturbance، measurement noise and random changes of model parameters. For more assessments، the muscle activation profile of involved muscles during the arm movement of an intact subject was compared with control signals obtained through the simulation studies. The results show an interesting timing synchronization between the activation and deactivation timing of control signals and muscle activation profiles.