Realization of Combined Motions by Robot Hand based on Surface and Deep Muscles' EMG

Abstract

We aim to realize combined motions of a robotic hand such as a myoelectric prosthetic arm by using Electromyography (EMG) of surface and deep muscles. A hybrid motion estimator is proposed to recognize hand motions corresponding to measured EMG and to estimate the joint angles during each hand motion. The hybrid motion estimator consists of Back-Propagation Neural Network (BPNN) and Multi-Input Single-Output (MISO) Nonlinear ARX (NARX) model. The hybrid motion estimator improve the estimation accuracy by considering a state transition from a previous state of hand motion to current one. The hybrid motion estimator has allowed to recognize both single motions, transition during single motions and a part of combined motions, and to estimate the corresponding joint angles with high accuracy. After verifying the effectiveness of the proposed estimator through numerical simulations, we have demonstrated that a robotic hand follows the estimated joint angles during recognized hand motion from measured surface and deep EMG of subjects.