Comparison of two data fusion methods for localization of wheeled mobile robot in farm conditions

Abstract

Localization of a mobile robot with any structure, work space and task is one of the most fundamental issues in the field of robotics and the prerequisite for moving any mobile robot that has always been a challenge for researchers. In this paper, the Dempster-Shafer (D.S.) and Kalman filter (K.F.) methods are used as the two main tools for the integration and processing of sensor data in robot localization to achieve the best estimate of positioning according to the unsteady environmental conditions in agricultural applications. Also, by providing a new method, the initial weighing on each of these GPS sensors and wheel encoders is done based on the reliability of each one. Also, using the two MAD and MSE criteria, the localization error was compared in both K.F. and D.S. methods. In normal Gaussian noise, the K.F. with a mean error of 2.59% performed better than the D.S. method with a 3.12% error. However, in terms of non-Gaussian noise exposure, the K.F. information was associated with a moderate error of 1.4, while the D.S. behavior in the face of these conditions was not significantly changed. The experimental tests confirmed the statement.