Optimization of EMG movement recognition for use in an upper limb wearable robot

Abstract

To functionally aid patients suffering from neurological disorder, a 3 degrees-of-freedom (DoF) upper limb wearable robot is presented (Fig. 1). In order to provide seamless user assistance, the intention of the wearer must be determined. As a sensing mechanism, electromyographic (EMG) signals have commonly been used to estimate human movement. In this study, the effectiveness of movement recognition using a generalized 8-port EMG sensor (Myo Armband) around the forearm was evaluated. Four fundamental movements of the arm (wrist flexion/extension and forearm pronation/supination) were classified using a neural network (NN) with a single hidden layer. The classification method was optimized through analysis of pre-processing algorithms and window size (0.25 to 1 second) to reduce computational expense and maintain classification accuracy. Through these accomplishments, significant groundwork has been provided for the development of a robust and non-invasive solution to tremor of the upper limb.