Simultaneous and Proportional Wrist Force Intent Estimation Method using Constrained Autoencoder

Abstract

The proposed model is capable of estimating multiple degrees of freedom wrist force simultaneously and proportionally. In the case of previous studies, it has been developed from a method of performing only a set motion based on pattern recognition, and studies that proportionally estimate wrist and finger strength are in progress. In this context, the wrist force was estimated using a method that modified the structure of the autoencoder, which is one of the artificial neural networks with high intention estimation performance compared to previous studies. This study was conducted with 2 degrees of freedom (Wrist flexion, Wrist extension, Ulnar deviation, Radial deviation) of the wrist. The proposed model showed high model accuracy and independence between degrees of freedom compared to similar previous models. Also, as a result of an online simulation test that reflects the real-time prosthetic control situation, 5 out of 6 performance indicators showed higher performance than the comparative model. As a result of this study, it was possible to infer that the accuracy of the model and the securing of independence between the degrees of freedom have a great effect on the control of the actual prosthetic arm including humans.