Rehabilitation exoskeleton is a wearable robot for recovery training of stroke patients. It is a complex human-robot interaction system with highly nonlinearities, such as modeling uncertainties, unknown human-robot interactive force, input constraints, and external disturbances. This paper focuses on trajectory tracking control of a rehabilitation exoskeleton knee joint which is driven by a hydraulic actuator with input saturation. A radial basis function neural network (RBF-NN) sliding mode repetitive learning control strategy is presented for the exoskeleton knee joint, where the RBF-NN is combined with a sliding mode surface to compensate for the modeling uncertainties and the controller difference as well as enhanced the robustness of the system. Incorporating with a nonlinear observer, a repetitive learning scheme is constructed to estimate the unknown external disturbances and learn the periodic human-robot interactive force caused by repetitive recovery training. Utilizing the Lyapunov approach, the stability of the closed-loop control system and the observer are guaranteed. Comparative simulation results verify the effectiveness of the proposed control scheme.
Rehabilitation exoskeleton is a wearable robot for recovery training of stroke patients. It is a complex human-robot interaction system with highly nonlinearities, such as modeling uncertainties, unknown human-robot interactive force, input constraints, and external disturbances. This paper focuses on trajectory tracking control of a rehabilitation exoskeleton knee joint which is driven by a hydraulic actuator with input saturation. A radial basis function neural network (RBF-NN) sliding mode repetitive learning control strategy is presented for the exoskeleton knee joint, where the RBF-NN is combined with a sliding mode surface to compensate for the modeling uncertainties and the controller difference as well as enhanced the robustness of the system. Incorporating with a nonlinear observer, a repetitive learning scheme is constructed to estimate the unknown external disturbances and learn the periodic human-robot interactive force caused by repetitive recovery training. Utilizing the Lyapunov approach, the stability of the closed-loop control system and the observer are guaranteed. Comparative simulation results verify the effectiveness of the proposed control scheme.