Contact force estimation for terminal traction human-machine interaction based on sEMG

In this paper, starting from the interaction force when the human body performs terminal traction, by comparing and selecting the most suitable sEMG features for upper limb terminal force estimation, and using muscle synergistic extraction to solve the problem of the multi-degree-of-freedom control mechanism of the upper limb, the long short-term memory neural network model (LSTM) is finally used to establish the contact force estimation model of the upper limb end. The root mean square errors of NMF-LSTM method in x, y and z directions are
$6.40\pm 0.39\mathrm{N},\, 5.14\pm 0.34\mathrm{N}$
and
$3.49\pm 0.20\mathrm{N}$
respectively. The average correlation coefficients in x, y and z directions are 0.93, 0.95 and 0.75 respectively, which is better than the results of PCA-LSTM, MLP and DT models.