A REAL-TIME EMG-DRIVEN ARM WRESTLING ROBOT CONSIDERING MOTION CHARACTERISTICS OF HUMAN UPPER LIMBS

Abstract

An EMG-driven Arm Wrestling Robot (AWR) is developed for the purposes of studying neuromuscular control of human elbow movements. The AWR arm has two degrees-of-freedom, integrated with mechanical arm, elbow/wrist force sensors, servo motors, encoders, MEMS accelerometers, and a USB camera, and is used to estimate tension generated by individual muscles from recorded electromyograms (EMG). The surface electromyography signal from the upper limb is sampled from a real player in the same conditions. By using the method of wavelet packet transformation (WPT) and autoregressive model (AR), the characteristics of EMG signals can be extracted. Then, an artificial neural network is adopted to estimate the elbow joint force. The effectiveness of the control method using force control estimated via neural network using WPT and AR as inputs is confirmed by experiments. The purpose of this paper is to describe the design objectives, fundamental components, and implementation of our real-time, EMG-driven AWR arm.