Extended Kalman Filter based fusion of reliable sensors using fuzzy logic

Abstract

Precise localization for autonomous robots is necessary for advancement in the world of unmanned robotics. Probabilistic algorithms are used to fuse multiple position sensors in order to locate a robot. But failure of any sensor in this process drastically lowers the performance of these algorithms. Here comes the need to facilitate these probabilistic models with intelligence. This paper presents an intelligent localization technique for autonomous maneuvering of robots. Localization of the robot is done by fusing three different types of position sensors using an Extended Kalman Filter (EKF) and a Kalman Filter (KF). The fusing method is made intelligent by keeping track of the relative error among the sensors and deciding a reliability factor on each sensor accordingly. A Fuzzy inference model has been adopted to predict the reliability factor for each sensor. According to the predicted reliability of each sensor, an error covariance matrix is set up, which is fed into the traditional KF and EKF algorithms. This helps the fusion algorithms to fuse the sensors intelligently and the final output is more accurate. A high precision localization is achieved by this intelligent method of fusing. A simulation is carried out in MATLAB considering three position sensors. The simulation is validated by making one of the sensors erroneous and comparing the output results of the new fusion algorithm with the traditional algorithm.