A Neural-Driven Musculoskeletal Model for Continuous Estimation of Hand and Wrist Movements

Abstract

Electromyography (EMG)-driven musculoskeletal models (MMs) have been developed for continuous control with multiple degrees of freedom. With the development of high-density surface EMG signal decomposition, extracted neural drive signals were widely used as inputs of the human-machine interface (HMI). This study proposed a neural-driven MM by using the neural drive signals as the inputs of the MM. The decomposed high-density EMG signals were used to extract motor unit (MU) discharge events and calculate the average discharge frequency to estimate the neural drive. The feasibility of the neural-driven MM in predicting the joint angles of the wrist and metacarpophalangeal (MCP) was tested on one ablebodied subject. The performance of the MM was evaluated by the Pearson's correlation coefficient (r) and the normalized root mean square error (NRMSE) between the estimated and measured joint angles. Compared with the EMG-driven MM, the neural-driven MM had higher r values and lower NRMSE values. The results demonstrated that the performance of the neural-driven MM was more accurate and robust with respect to that of the EMG-driven MM. The proposed neural-driven MM has great potential for the robust continuous control of wearable robots.