Grasping force prediction based on sEMG signals

In order to realize the force control, when the prosthetic hand grasps the object, the forearm electromyography signal is collected by the multi-channel surface electromyography (sEMG) acquisition system. The grasping force information of the human hand is recorded by the six-dimensional force sensor. The root mean square (RMS) of the electromyography signal steady state is selected, which is an effective feature. The gene expression programming algorithm (GEP) and BP neural network are used to construct the prediction model and predict the grasping force. The force prediction accuracy of GEP algorithm and BP neural network algorithm are discussed under different grasping power levels and different grasping modes. The performance of the two algorithm models are evaluated by two measures of root mean square error (RMSE) and correlation coefficient (CC). The results show that the RMS eigenvalue extracted from the sEMG signal can better characterize the grasping force. The prediction model with GEP algorithm has smaller relative error and higher prediction effect