Machine Learning for Human-Robot Interaction using EMG Signals to Control the Grip Force

This paper presents a strategy for human-robot interaction (HRI) utilizing electromyography (EMG) signals. The main aim is to enable a robotic manipulator to receive a desired gripe force during handling an object according to muscular activation of its user. To this end, we exploit from Support Vector Machine (SVM) as a powerful machine learning tool for classification. The EMG signals have been recorded from 3 major muscles of the forearm acquired by 4 participants. Each one was asked to squeeze the dynamometer handle which measuring their grip force, while the force was 0 (rest condition) as well as it was steadied at 3 desired values of 5, 10 and 20 kg, the EMG signals were recorded for 5 seconds in each case. These signals are used to build a model that classifies the grip force which must be applied by a robot gripper based on the user muscular activity. After building the SVM model, it is examined to control the grip force of a simulated robotic manipulator during transmitting an object using CoppeliaSim software. Results indicate the successful performance of this strategy.