XGB ASSISTED SELF-LEARNING KALMAN FILTER FOR UWB LOCALIZATION

Abstract

Recent years, more and more mobile robot has been used in many fields. In order to improve the service quality of mobile robot, how to improve the accuracy of robot position information has gradually become a research hotspot in this field. In this work, we will focus on the following situation: in an indoor environment, one mobile robot moves along one similar trajectory repeatedly. And the extreme gradient boosting (XGB) assisted self-learning Kalman filter (KF) will be derived in this work. To the method, the XGB is used to build the mapping between the distances from the ultra wide band (UWB) reference nodes (RNs) to the UWB blind node (BN) and the mobile robot’s position. Then, the XGB is used to build the measurement of the Kalman filter by using the off-line and on-line mode, which is able to provide the accurate position information. The real test has bee done, and the results show that the proposed XGB assisted self-learning KF is able to improve the localization accuracy gradually.