Manufacturing and Controlling Robotic Arm Through Machine Learning via Electromyography Signal

Abstract

Industrial robotics has advanced significantly as a result of automation's quick progress, and its applications in biomedicine, particularly in rehabilitation technology are growing. The goal of this project is to design, construct, manufacture and test a low-cost, six-degree-of-freedom 3D-printed Thor robotic arm that is controlled by surface electromyography (sEMG) signals. These signals were gathered from nine healthy volunteers utilizing a wireless Myo motions armband, and they made it possible to distinguish between seven different hand gestures.

Because of its mechanical robustness, polylactic acid (PLA) was employed in 3D printing, which added to the Thor prototype's endurance and open-source nature. Two Arduino Mega controllers and Ramps interface boards improved the electronic system, and Marlin firmware and Arduino-based software were used to govern control. Workspace navigation was used to evaluate the system and install an extra stepper motor to increase torque at the base joint.

Support Vector Machines (SVM), K-Nearest Neighbors (K-NN), and Linear Discriminant Analysis (LDA) were the three classifiers used for gesture detection. With an accuracy of 95.04% for dominant hands and 83.41% for non-dominant hands, SVM demonstrated the best accuracy. Additionally, SVM achieved near-perfect accuracy in elbow flexion-extension identification (99.55% for dominant and 98.73% for non-dominant), while LDA and K-NN also shown strong performance.