Modelling and Control of a Robotic Arm Using Artificial Neural Network

Abstract

Often it can be seen that men with a lost arm face severe difficulties doing daily chores. Artificial Intelligence could be effectively used to provide some respite to those people. Neural networks and their applications have been an active research topic since recent past in the rehabilitation robotics/machine learning community, as it can be used to predict posture/gesture which is guided by signals from the human brain. In this paper, a method is proposed to estimate force from Surface Electromyography (s-EMG) signals generated by specific hand movements and then design and control a Robotic arm using Artificial Neural Network (ANN) to replicate human arm. Here the force prediction is a Regression process. A hand model has been successfully moved using servo motor that has been programmed based on the results obtained from sample data. The results shown in this paper illustrate how the Robotic arm performs. Index Terms: Surface EMG, Artificial Neural Network, Robotic arm, Regression. I. Introduction This project has been designed on data taken as Surface EMG signals from the human arm. Surface Electromyography is a non-invasive technique for measuring muscle electrical activity that occurs during mus- cle contraction and relaxation cycles. EMG signals contain the information about the muscle force which can be used in human-machine interaction. Force plays an important role in these applications. Rezazadeh et al. (1) proposed a co-adaptive Human-Machine Interface (HMI) that is developed to control virtual forearm prosthesis over a long period of operation. This paper has influenced us to make a robotic arm based on the hand EMG signal. The physical structure of the robotic arm has been modeled such that it can be easily assembled. It has been simplified such that it has 1 degree of freedom (with capability of rotating 180 ◦ ) while retaining some of the important motions of the human arm. A five-finger model has been constructed. The working is based on the simple fact of comparison of performances of EMG signals taken from human arm and the target to be achieved. This model exhibits the full range of motion required to move the arm with some degree of force. A virtual re- ality model package has been developed in MATLAB to control the model in such a manner that it is able to demonstrate the potential of the work. Mobasser et al. (2) used multilayer perceptron ANN for hand force esti- mation from surface EMG signals for applications in sports activities. Yang et al. (3) demonstrated the use of ANN, Locally Weighted Projection Regression (LWPR) and SVM to estimate hand grasp force from surface EMG signals for force control of multi-functional prosthetic hands. Haritha et al. (4) used a method to estimate the hand force from Surface Electromyography signals using ANN.In this work, we will estimate the predicted value of force from the surface EMG signals using a feed-forward ANN. The neural network is trained with both EMG and force data. Then the arm model is moved using servo motor which is programmed based on the results obtained from sample data.The next section describes the hardware setup of our system. Section III pre- sents the methodology we proposed for force estimation and operation of the robotic arm. Section IV discusses the experimental evaluation and results. Section V presents the conclusion of the paper.