Manipulation of Robot Hand Based on Motion Estimation Using EMG Signals

Abstract

The purpose of this research is the construction of an intelligence artificial arm control system which uses EMG signals. Signal processings of EMG signals are preformed using multiple regression equation and Mahalanobis' distance, which can learn parameters in a short time. The motion pattern discrimination is conducted with both technique and joint angles are predicted with multiple regression equation. A multiple regression model provides a less adequate accuracy than an artificial neural network generally used, but accuracy is improved by selection of suitable inputs and generation of teacher signals. Joint angles were predicted from joint angles data of the past and the signals that extracted characteristics of each motion. The experiments were conducted to verify the validity of this technique. Discriminated motions were grip, open and chuck at a hand. Predicted joint angles were multi-finger angles corresponding to these three motions. First, we verified these motions are discriminable from EMG signals, and the motion pattern discrimination was conducted. Consequently, these motions were discriminated. Second, the experiment using a robot hand was conducted. A robot hand was controlled intuitively and accurately in all five subjects. From these experiments, the usefulness of processing EMG signals with proposed methods were proved.