Localization of an indoor mobile robot using decentralized data fusion

Abstract

An important problem in navigation of a mobile robot is localization. The robot can estimate its location in a predefined environment by combining the information obtained from its sensors. There are various methods to combine the information obtained from sensors and the models used to estimate the position and velocity of the robot. In this paper, a decentralized data fusion algorithm is used, where the sensors are used in several independent estimation loops. Then, the estimates of these loops are exchanged to improve the accuracy of the position estimates. The sensors used in this study consist of a laser scanner, Kinect depth sensor, encoder and IMU, which are divided into two local loops. The algorithm was implemented in a real environment on a robot built at the Mechatronics Laboratory of the Department of Mechanical Engineering, Yazd University. It is observed that the decentralized estimates of the two loops track the real trajectory of the robot satisfactorily. It is observed that the precision of the decentralized approach is somewhat less than the accuracy of the centralized data fusion, since different kinematic models are used in the two loops and also due to asynchrony of the sensors data. However, the decentralized approach entails advantages such as robustness of estimation, possibility of error crosschecking and sensor failure detection.