A novel method based on long short term memory network and discrete-time zeroing neural algorithm for upper-limb continuous estimation using sEMG signals

Abstract

In this paper, a novel closed-loop model based on surface electromyography (sEMG) comprised a long short term memory (LSTM) network and discrete-time zeroing neural algorithm called zeroing neural network (ZNN), which is developed to estimate joint angles and angular velocities of human upper limb with joint damping. The dynamic model of human upper limb with joint damping is set up as the initial equation. Then, the LSTM network is proposed as an open-loop model which described the input-output relationship between the sEMG signals and joint motion intention. Besides, a novel closed-loop model is built via ZNN for eliminating the predicted error of open-loop model and improving the accuracy of motion intention recognition. Founded on the sEMG signals, the continuous movement of human upper limb joint can be successfully estimated via the novel closed-loop model. The results show that for simple joint movements, the closed-loop model is able to estimate the movement intention of human upper limb with high accuracy.