An Online sEMG Motion Classification Framework for Tele-operating the Robotic Hand

Abstract

Seamless communication between human intended motions and robot actions is essential for Human-Robot Interaction (HRI). When tele-operating a robotic hand, it is a natural and effective way via surface electromyography (sEMG) signals. This paper proposes an online sEMG motion classification framework for tele-operating the robotic hand. The whole framework consists of offline training and online recognition phases. In the offline training phase, three features were selected from four candidates and Artificial Neural Network (ANN) won the election among three classifiers. Inthe online recognition phase, two-thresholds data segmentation and majority voting techniques were designed, and three subjects participated in online experiment to verify the feasibility of this framework. The online experimental results show that the average total accuracy is 73.56% and the average vote accuracy is 91.67%. The outcomes of this study have shown the promising potential of sEMG-based HRI.