Continuous Finger Kinematics Estimation Based on sEMG and Attention-ConvGRU Network

sEMG is an efficient media for human-computer interactions, especially in the controlling of artificial limbs and other mechanical arms. In the paper, we propose a new Attention-ConvGRU model which can continuously estimate the finger joint angles based on sEMG when executing classical grasping motions. The experimental results show that the average correlation coefficient (CC) and the root mean square error (RMSE) of the proposed Attention-ConvGRU method are 0.8320 +/- 0.04 and 8.8717 +/- 0.98, respectively, which are significantly better than that of the GRU method (0.8093 +/- 0.05, p < 0.01, 9.3716 +/- 0.95, p < 0.01). Moreover, the training speed of Attention-ConvGRU is 4.5 times faster than that of GRU method.