Simultaneous and Proportional Estimation of Multijoint Kinematics From EMG Signals for Myocontrol of Robotic Hands

As the primary effector of human, hand is dexterous, important, and biomechanically complex. How to control the robotic hand as human intends in the human-centered robotic systems is a hot topic to guarantee natural interaction between users and robots. In this article, we investigate the simultaneous and proportional estimation of human's movement intent from surface electromyography (EMG) signals. A sparse pseudo-input Gaussian process regression method is proposed to map the EMG features and the hand kinematics. The kinematics of five primary degrees of freedom (DoFs) are estimated from EMG recorded during functional grasping tasks. From the experiments on eight able-bodied subjects, ipsi-lateral and contra-lateral training strategies represent similar estimation accuracy (CC = 0.89 and 0.88, respectively) and no significant difference ( p = 0.8) between them. The online estimation with contra-lateral strategy shows higher average estimation accuracy (CC = 0.91). In addition, the estimation of the intended kinematics can be decoded in real time with negligible response delays. It is suggested to apply the proposed method to multi-DoF decoding for natural, intuitive, and accurate myocontrol of robotic hands.