A Novel Wrist Joint Torque Prediction Method Based on EMG and LSTM

Abstract

Electromyography (EMG) signal is one of biological signals, it has been used in rehabilitation and prosthesis for several decades. For the control of prosthetic hands, EMG signals are usually used to make classification to recognize hand gestures or wrist motion patterns, but most research neglect the continuous motion of wrist, as a result, those prosthetic hands can't move continuously as user's intention. In this study, we proposed a novel method based on long short-term memory (LSTM) to predict continuous wrist joint torque from EMG signal. Five healthy subjects participated in the experiment, subjects were asked to flex or extend their wrist with load in their hand to simulate daily scene in life, at the same time, wrist joint angle and EMG signals were recorded synchronously. Then filtering the raw EMG to get useful components, and three time-domain features were extracted from the pre-processed EMG signals, after that reference wrist joint torque were calculated from the wrist joint motion dynamic model. At last, a wrist joint torque prediction model was established based on LSTM, and Pearson correlation coefficient between the predicted result and reference wrist joint torque was used to evaluate the performance of the proposed method. The Pearson correlation coefficient of all subjects range from 0.8669 to 0.9772, and 0.9289 on average which means the proposed wrist torque prediction model can well predict wrist torque. The proposed method can be used in the control of prosthetic limb to move continuously.