Movement Classification for Hand Telerobot Based on Electromyography Signal Using Convolutional Neural Networks

Abstract

A telerobot or teleoperated robot is a device that is remotely controlled by a human operator as opposed to following a predetermined sequence of movements, but which exhibits semi-autonomous behavior [1] Combining the major subfields of teleoperation and telepresence. In the scientific and technical community, teleoperation is the most common term for remote operation. Telepresence, on the other hand, is a subset of telerobotic systems equipped with an immersive interface that allows the operator to feel physically present in a remote environment by conveying his presence through the remote robot [2]. In a teleoperation system, there are two main components that must be prepared. Namely the controller or robot operator (local site) and the controlled robot (remote site) [3]. Telerobot requires a transmission medium to communicate. WebSocket is very appropriate to be used as a telerobot communication protocol [4]. There are many types of telerobot, including hand robot. In this study, a telerobot uses the convolutional neural network (CNN) algorithm to classify hand movements based on electromyography signal. Model testing has a training process and model tuning to get the best hyperparameter value. These results are tested on five subjects and call an average accuracy of 0.996600, f1 score of 0.996634, and precision of 0.996309 with a data composition of 80% for training data and 20% for testing data. To test the telerobot time delay by sending several motion codes from the local site to the gateway server and forwarded to the remote site with a local site bandwidth of 7.79 Mbps download and 1.56 Mbps upload and a remote site bandwidth of 11.6 Mbps for download and 5.96 Mbps for upload, the resulting average values average of 0.395 seconds or 395 ms.