Mobile robot localization using fuzzy neural network based extended Kalman filter

Abstract

Localization is fundamental to autonomous operation of the mobile robot. In this paper, a new optimal filter namely fuzzy neural network based extended Kalman filter (FNN-EKF) is introduced to improve the localization of a mobile robot in unknown environment. The filter is a combination between a normal extended Kalman filter (EKF) installed on a differential-drive wheeled mobile robot and an online adjustment of the process noise covariance matrix Q and the measurement noise covariance matrix R. The adjustment is performed by fuzzy system and the purpose is to overcome the divergence of the EKF when the matrices Q and R are fixed or wrongly determined. The membership functions of the antecedent and consequent parts of fuzzy if-then rules in the fuzzy system are tuned by neural network. Integrating neural network into the fuzzy system called the fuzzy neural network is to gain the accuracy while reducing the time and cost in designing the membership functions. Simulating experiments have been conducted and results show that the FNN - EKF is more accurate than the EKF in localizing the mobile robot. An evaluation of the system with respect to suggestions of possible future developments is also mentioned in the paper.