Continuous And Simultaneous Emg-Based Neural Network Control Of Transradial Prostheses

Abstract

As the development of dexterous prosthetic hand and wrist units continues, there is a need for command interfaces that will enable a user to operate these multi-joint devices in a natural, coordinated manner. In this study, myoelectric signals and hand kinematics were recorded as three able-bodied subjects performed a variety of individuated movements and simulated functional tasks. Time-delayed artificial neural networks (TDANNs) were designed to simultaneously decode the movement trajectories for seven distal degrees of freedom (pronation-supination, wrist ulnar-radial deviation, wrist flexion-extension, thumb rotation, thumb abduction-adduction, finger MCP flexion-extension, and finger PIP flexion-extension). Performance was quantified by calculating the variance accounted for (VAF) and normalized root-mean-square error (NRMSE) between the decoded and actual movements. Accurate predictions were achieved (VAF: 0.57-0.80, NRMSE: 0.04-0.11), suggesting that it may be possible to provide an intuitive EMG-based scheme that provides continuous and simultaneous multi-joint control for individuals with below-elbow amputations.