Design of Elbow Rehabilitation Exoskeleton Robot with sEMG-based Torque Estimation Control Strategy

Abstract

The surface electromyography (sEMG)-based human joint torque estimation algorithm has shown a great potential in the human-computer interaction function. Human elbow joint requires different forearm postures to accomplish specific tasks while flexing the forearm, such as grasping a tabletop object in the pronation (Pro) position, picking up a glass of water in the neutral (Neu) position, and lifting a heavy object in the supination (Sup) position. Existing elbow torque estimation algorithms only capture single channel sEMG signals in a fixed forearm posture, while ignore the effects of muscle force changes during elbow flexion in other postures. In this paper, we designed an elbow exoskeleton and developed a backpropagation neural network (BPNN)-based joint torque regression algorithm for rehabilitation. Specifically, four-channel sEMG data from eight subjects in three different forearm postures with elbow flexion were collected. An assistive rehabilitation model based on the BPNN algorithm for elbow was trained. Experimental results showed that our proposed approach reduced muscle activation by an average of 34.53±7.48% in different postures, and reduced the human active torque by 66.45±8.37%. The superior performance indicated that the torque estimation algorithm and control strategy proposed in this study can improve the generalizability and practicality of rehabilitation robots.