Bio-Signal Controlled Exoskeleton for Accurate Elbow Rehabilitation

Abstract

To address the limitations of current elbow rehabilitation exoskeletons—namely insufficient force-control accuracy, low active user engagement, and poor adaptability to individual needs—this work develops a precision-controlled exoskeleton system based on biological signal perception. The system acquires muscle activity through surface electromyography (sEMG) sensors and identifies movement intentions using a convolutional neural network (CNN). A differential transmission mechanism is employed to drive elbow flexion–extension and forearm pronation–supination, enabling two degrees of freedom. Experiments were conducted with four healthy subjects to compare sEMG signals (RMS) and primary joint torque between a free mode and an assistive mode under three forearm postures (neutral, pronation, and supination). Results show that, in assistive mode, the RMS of major upper-limb muscles decreased by an average of 38.53 ± 11.98%, and primary joint torque decreased by an average of 58.5 ± 17.27%. The brachioradialis exhibited the largest RMS reduction in the neutral posture (54.76 ± 16.69%), while the supination posture resulted in the greatest torque reduction (66.2 ± 15.34%). These findings demonstrate that the proposed system effectively reduces muscle activation and voluntary effort, validating its effectiveness in rehabilitation training and the stability of human–robot cooperative control.