Neural adaptive assist-as-needed control for rehabilitation robots

Abstract

© 2018 Australasian Robotics and Automation Association. All rights reserved. Robot-assisted therapy can improve motor function in patients recovering from stroke. Assist-as-needed algorithms provide only minimal robotic assistance in the therapy, thus requiring significant effort from the impaired subject. This paper presents an adaptive neural assist-as-needed controller for rehabilitative robots. The controller combines the Lyapunov direct method with the computed torque control and neural networks. Robot assistance is limited to only as needed by adding the force reducing term into the adaptive control law. This paper shows that by the presented method the tracking error converges to a small value around zero while the neural network weights and system uncertainties remain bounded. Simulation on a robot manipulator model is presented to demonstrate the effectiveness of the proposed meth.