Estimation of elbow joint angle by NARX model using EMG data

Myoelectric control has a key role in human-machine interface applications such as orthosis control and teleoperation. Myoelectric signals are bio signals that are detectable from surface of the skin, and contain useful information about user's moving intention. This paper presents a methodology to estimate elbow joint angle from muscle's data using neural network (NN). Proposed methodology can be expanded to estimate any joint angle by recording muscle activities concerning with the joint. In addition, Nonlinear Auto-Regressive eXogenous-NN (NARX-NN) model is selected to estimate joint angle. Several data sets are recorded and processed to train and test the NN. The trained network is used to predict elbow angles for individual data sets. Results show that trained network can estimate joint angle with an acceptable performance.