Real-time Multiple-Channel Shoulder EMG Processing for a Rehabilitative Upper-limb Exoskeleton Motion Control Using ANN Machine Learning

Abstract

Machine learning (ML) has been proposed for surface electromyography (sEMG) signal processing to control assistive robots. However, most of the studies using ML focused on recording sEMG for off-line data analysis and pattern recognition. Only a few studies reported online ML EMG processing for one DoF hand motion pattern recognition and instant robot control. Using ML for multiple channels of shoulder EMG signal processing and robotic motion control has not been reported in literature. In this paper, we report the efficiency of a 12-channel shoulder sEMG signal processing system using ML techniques for instant upper limb exoskeleton motion control. The Delsys EMG sensor system, LabVIEW program with ML toolbox, and an upper limb exoskeleton were used together with DAQ and Arduino boards for exoskeleton motion control. Artificial neural-network (ANN) ML technique was used to recognize shoulder motion patterns. The accuracy of motion pattern recognition was validated among eighteen subjects. The experimental results showed that the average accuracy of shoulder motion pattern recognition was 87.98% - 99.34% in offline validation and 74.0 - 98.0% in instant online testing, respectively. This study demonstrated an approach of real-time multiple-channel shoulder sEMG signals processing for upper limb exoskeleton motion control using ANN ML.