SVM Classification of EMG Signals for Control of a Robotic Hand

Over the last decades, research on robotic devices for the replacement of lost limbs, such as actively actuated prostheses, has been intensified. In this paper we present the development of a low-cost robotic arm together with a sensor bracelet for electromyographic (EMG) control. With a total cost of about $300 the introduced system aims at contributing to the development of inexpensive alternatives for commercially available products. The muscular activity measured on two positions on the forearm is used to recognize pre-defined hand motions, or more precisely gestures and grasps. To this end, the machine learning method support vector machines (SVM) is used as classification algorithm, achieving a mean classification accuracy of 89.45%, with a minimum accuracy of 85.63% and a maximum accuracy of 98.02% for the individual motions. The recognized gestures and grasps are then executed on the robotic device, whereby all five fingers of the hand segment are actuated with servo motors and cables.