EMG Based Control of Individual Fingers of Robotic Hand

Abstract

Research on robotic hand for prosthesis is widely being carried out. This paper attempts to provide an enhanced control of individual fingers of robotic hand using electromyography (EMG). Eight-channel surface EMG is to acquire signals from widest part of forearm of 10 subjects. These signals are corresponding to movement of five individual fingers, which are thumb, index finger, middle finger, ring finger, and little finger. Features are extracted from these signals and Deep Neural Network is applied as classifier to distinguish between the five signals with an average accuracy of around 95%. The classified signal is then interfaced with the robotic hand to perform mimicking of individual finger motion. This work can be effectively used for hand rehabilitation with flexibility of actively controlling individual finger motion.