EMG-based estimation of shoulder and elbow joint angles for intuitive myoelectric control

Surface Electromyography (EMG) has been considered as one of the modalities of human-machine interface (HMI) in the context of human-centered robotics. In order to interpret human muscle activities into motion intents, pattern classification-based EMG decoding methods and continuous joint kinematics methods were both proposed for advanced motion control. The former mainly provided binary command to activate a single DoF or a predefined motion pattern at one time, while the latter mainly estimated the joint kinematics of individual motion. In this work, we proposed to take advantage of these two technologies to achieve intuitive joint angle estimation for multiple arm motions concurrently. That is, the result of motion classification was applied to select correct joint angle estimation artificial neural network (ANN) which was trained for each motion in advance. Principal component analysis (PCA) processing presented its contribution to the improvement of the motion classification accuracy. The motion classification accuracy is around 92% across four subjects with least-square support vector machine (LS-SVM). The joint angle estimation represents around 80% accuracy of four arm motions across four subjects. This result indicates that the proposed method with the combination of the pattern classification and concurrent joint angle estimation is viable and promising to be applied for intuitive myoelectric control.