Artificial neural networks for estimation of joint angle from EMG signals

Abstract

A set of neural networks was developed for EMG based control of telemanipulators. The neural network system provides an estimate of the joint angle at the elbow using surface EMG of biceps in real time. The joint angle was measured by a goniometer to calibrate the system and train the networks. Preliminary results during testing indicate an error of less than 20% between the joint angle estimate of the network and the actual joint angle.