Robot Trajectory Target Delivery Using Machine Learning

Abstract

People who have lost their arm will most likely face many issues performing tasks that those with their arms would consider simplistic. If they have lost their arm while they are older, these people now face issues that they have never experienced with their newfound mobility restrictions. Currently, artificial limbs that have been controlled by electromagnetic signals that are classified by artificial neural networks have had the best success in solving this issue, as supported by previous research.In this paper, we will discuss the simulation of a human arm using an open source software called OpenSim to model a biologically accurate human arm and implement a deep learning program using a long short-term memory deep neural network created in MATLAB. As a proof of concept, we present that the arm is able to use the deep learning network to toss a 0.25 kg ball onto a specified target location accurately.Our work shows that a model of an artificial arm can learn to do simple tasks, such as tossing a ball the correct distance, using a long short-term memory deep neural network, proving that machine learning and deep learning have an effective application in prosthetic and robotic arm design.