Motion Compensation of Omnidirectional Wheel Robot Using Neural Networks

Abstract

This paper describes a method to compensate the omnidirectional robot's motion control. Due to the variation of individual mechanical characteristics, there always have some errors between the given command and the robot's execution. The paper explains how we cope with the execution errors by two neural networks modeling the rotation and translation individually. And we have successfully field-tests the compensation method at several RoboCup events with our ZjuNlict team. The result shows that the compensation can significantly improve the accuracy of robotic play