Continuous Estimation of Grasp Kinematics with Real-Time Surface EMG Decomposition

Abstract

The aim of the study was to apply the real-time surface electromyography (EMG) decomposition to the continuous estimation of grasp kinematics. A real-time decomposition scheme based on the convolutional compensation kernel algorithm was proposed. High-density surface EMG signals and grasp kinematics were recorded concurrently from five able-bodied subject. The electro-mechanical delay between identified motor unit activities and grasp kinematics was characterized and utilized to optimize the multiple linear regression model for the grasp estimation. The discharge rate of each motor unit was extracted as the feature input to the regression model. On average, \(36\pm 15\) motor units were identified during each grasp task. The average root mean square error between estimated grasp kinematics and actual recorded signals was \(0.21\pm 0.05\), with the average delay of \(212\pm 50\) ms for the feature. The computation efficiency of the decomposition scheme and the high estimation accuracy imply the practical application for human-machine interfaces based on neural signals.