Sparse Gaussian process regression in real-time myoelectric control

Myong Chol Jung, Rifai Chai, Jinchuan Zheng, Hung Tan Nguyen

Abstract: In myoelectric control, nonlinear regression models, Gaussian process (GP) in specific, have shown promising accuracy in estimation, but no study has been conducted to evaluate the real-time performance of GP regression. In this work, the real-time performance of sparse GP regression is evaluated with 17 able-bodied subjects. Unlike the existing training methods, in which training protocols are strictly pre-determined, a novel training method is proposed. The subjects' real-time performance adjusts training time and the number of training samples. While the majority of subjects showed similar learning rates, there was a significant difference between a few subjects (p < 0.05). As a result of real-time performance, the subjects completed 97% of the average tasks and achieved 80% path efficiency comparable to existing methods.