STUDY OF FRICTION COMPENSATION MODEL FOR MOBILE ROBOT’S JOINTS

Abstract

Frictional forces inside the joints of mobile robots hurt robot operation's stability and positioning accuracy. Therefore, establishing a suitable friction force compensation model has been a hot research topic in robotics. To explore the robot joint friction compensation model, three friction compensation models: linear, nonlinear, and neural network models, are developed in this paper. Based on the deep learning algorithm for three models at low speed, high speed, acceleration, and uniform speed training test, respectively results have been obtained. The test results show that the best friction compensation effect comes from combining neural network models in acceleration and a consistent speed state way. The friction compensation model trained this way yielded superior results to the other combinations tested. Finally, using the method, a friction compensation model trained by adding a neural network to the feedforward control torque was tested on a four-wheeled mobile robot platform. The test results show that the relative error of the torque caused by the friction of each joint is reduced by 15%-75% in 8 groups of tests, which indicates that our friction compensation method has a positive effect on improving the accuracy of the joint torque.