Reinforcement Learning Adaptive Control for Upper Limb Rehabilitation Robot Based on Fuzzy Neural Network

Aiming to how to coordinate and control the patient's upper limb to trace the set train motion trajectory and position which are purposed base on the statues of the sick upper limb, the paper purposed a novel reinforcement leaning controller. In the continuous-time RL scheme, a fuzzy actor is employed to approximate the plant(which includes rehabilitation robot and the sick upper-limb), and a critic NN is designed to evaluate the performance of the actor. At the same time, the critic NN generates some rewards back to the fuzzy actor for tuning weight of rules. The weight tuning law is given based on Lyapunov stability analysis. The purposed RL was finally simulated and analyzed, experiment and simulation results showed that the control strategy not only effectively provided the robot's tracking requirements, but also had strong robustness and flexibility.