User independent hand motion recognition for robot arm manipulation

Abstract

In the present work, tele-manipulation of robot arm and gripper is experimentally performed using inertia measurement unit (IMU) and electromyogram (EMG)-based human motion recognition. The movement of robot arm and motion of robot gripper is determined based on the measured IMU and EMG data, respectively. To overcome user dependence which is one of main disadvantage of EMG-based motion recognition, reference voluntary contraction method-based normalization of measured EMG data is carried out. Training and test data of EMG are obtained from experiments for four kinds of hand motion of four experimental participants. After extraction of feature vectors, artificial neural network is applied for the EMG-based hand motion recognition. Even when training data and test data are obtained from different participants, it is confirmed that classification accuracy can be greatly improved through the proposed simple normalization method. Finally, a real-time tele-manipulation of 6-degree-of-freedom robot arm is demonstrated successfully by adopting the proposed user independent human motion recognition method.