SEMG-based decoding of detailed human intentions from finger-level hand motions

Abstract

In this paper, a novel sEMG decoder was proposed for estimating the detailed human intentions. For the use in practical applications, an sEMG decoder needs to have a high decoding performance, a robustness to the changes of electrode positions, and a capability to extract detailed and continuous intentions behind human motions. The new decoder satisfies all these needs by using a new supervised feature extractor, which was originally designed for discrete output values but modified to deal with the continuous values while maintaining the high performance and the robustness to the change of input variables resulting from the change of electrode positions. Experiments for decoding detailed human intentions, i.e. the angle between phalanges in a single finger, confirms that the high performance of the proposed decoder and that this performance is maintained when the electrode position changes.