Electromyographic Driven Assisted Therapy for Hand Rehabilitation by Robotic Orthosis and Artificial Neural Networks

Abstract

This study addressed the design, construction and instrumentation of an active hand orthosis driven by electromyographic (EMG) signals captured over the arm. The EMG signals were classified using a static neural network (SNN) supplied by the momentum learning scheme. The orthosis was actuated with a collection of direct current motor commanded by distributed control strategy based on the so-called twisting controller. The orthosis was interfaced to a computer where a special class of graphic user interface (GUI) was implemented. This GUI contains a sequence of suggested exercises that patient wearing the orthosis must try to develop. The orthosis was implemented and the proposed controller forced the tracking of the reference trajectories supplied by the GUI. The orthosis was evaluated in simulation to adjust the EMG signal classifier as well as the controller gains. A real orthosis was constructed and controlled using the gains obtained at the simulation stage.