Finger joint angle estimation based on sEMG signals by Attention-MLP

Abstract

sEMG(Surface electromyography) signal was widely applied in human-machine interactive field, especially in robotic arm control. In this paper, we built the Attention-MLP (Multilayer Perceptron) model to implement a type of continuous joint angle estimation method based on sEMG for six grasp movements, we tested this model on Ninapro dataset and the average Pearson correlation coefficient (CC) and the average root mean square error (RMSE) of the proposed Attention-MLP method achieved 0.812±0.02 and 10.51±1.98; the average CC and RMSE of this method are better than Sparse Pseudo-input Gaussian processes (SPGP), its average CC and RMSE are 12.14±2.30 and 0.727±0.07. Compared with the traditional method SPGP, our model performed better on continuously estimation of ten main hand joint angles under 6 grip movements.