Three-dimensional simultaneous EMG control based on multi-layer support vector regression with interactive structure

In this paper, a novel three-dimensional (3D) simultaneous myoelectric (electromyography, EMG) control scheme established on multiple layers of support vector regression (SVR) with an interactive structure was proposed. For choosing a proper set of the three degrees of freedom (DOF’s), a variety of DOF combinations (e.g., flexion/extension of the thumb and fingers, wrist pronation/ supination, etc.) were compared in terms of their regression accuracy. The effort to drive a particular DOF for achieving a given motion with a specific strength was quantified through the root mean square (RMS) of the multichannel signals, and then used to train a three-layer SVR model. An interactive structure was specially introduced in the model for improving the learning efficiency and control performance by taking advantage of the prior, supplementary regression knowledge. Both offline evaluation (regression criterions) and online experiments (3D target positioning) were conducted to verify our method’s efficacy.