Manipulation of a robot by EMG signals using linear multiple regression model

Abstract

In this research, a robot was operated by EMG signals using a linear multiple regression model. Myoelectric upper-limb prostheses are one example of an application that employs EMG signals as a control input. However, commercial myoelectric upper-limb prostheses can perform only grasping motions and wrist rotation. Many researches on multifunctionalization of myoelectric upper-limb prostheses have been undertaken, and pattern recognition for discriminating desired motions of hands from EMG signals have been attempted. Artificial neural networks are commonly applied in these cases. Since EMG signals have nonlinear characteristics, it is reasonable to use artificial neural networks to produce accurate nonlinear maps. However, this is not practical because large amounts of training time are necessary before actual use. In this research, signals that predict operation using our linear multiple regression models are generated, and although a learning process is also needed in this method, it takes only a short time. Using this technique, we were able to discern forearm motion and predict an elbow joint angle. The usefulness was verified by an experiment using a robot hand and a robot arm.