EMG-based Control Scheme with SVM Classifier for Assistive Robot Arm

Establishing an appropriate control technique with high accuracy has become the central issue for robotic applications to achieve anthropomorphic functions. The electromyography (EMG) method, which approaches various human motions by adjusting signals in the manipulator control scheme, is especially gaining popularity among assistive robotics. In this study, an EMG-based classification method is incorporated to control an assistive robot arm and perform interaction in human arm movements. The system is implemented based on a support-vector machine (SVM); it classifies motions of upper human limbs according to EMG signals recorded from the brachioradialis, the biceps and the anterior deltoid as the input features. With the proposed method, six categories of human arm movements can be identified, and the classification results can thus be remotely transmitted to control an assistive robot to mimic motions of the human arm. The performance of the proposed classification method was evaluated using 72 segments in 5 second EMG signals from two subject. The overall accuracy rate can reach 94% based on the selected EMG features for each measured muscle. The results suggest the importance of feature selection according to the morphology of EMG waveforms recorded from different muscle bundles. The performance of the proposed method has been quantitatively evaluated, and an assistive robot arm can be remotely controlled using EMG signals with a high accuracy rate.