EMG-based Continuous Prediction of the Upper Limb Elbow Joint Angle Using GRNN

Electromyography (EMG) signal is one of the important applied biological signals generated by human muscles which represents the human motor intention. In this paper, we proposed a prediction model of the elbow joint angle in the sagittal plane only using the EMG signal. After the signal acquisition and the pre-processing, the root mean square (RMS) feature of EMG signal was extracted as the input set of the General Regression Neural Network (GRNN) which has a high performance in the nonlinear fitting. Furthermore, three general regression neural networks were set up to predict the elbow angle in raising step, holding step and falling step, respectively. To evaluated the efficiency of the method, the subjects were asked to flex, hold and extend their forearm continuously in every 3 seconds. Meanwhile, the real angle and the predicted angle were compared in the experiment. The experimental results indicated the proposed method can predict the elbow angle in the sagittal plane with the best error rate no more than 5%.