Multiple sensor based UGV localization using fuzzy extended Kalman filtering

Abstract

A method for localization of unmanned ground vehicles (UGVs) that are equipped with multiple sensors is presented and evaluated based on derivation of a fuzzy extended Kalman filter (EKF). The fuzzy EKF is used to fuse information acquired from the UGV odometer, stereo vision system and laser range finder in order to estimate the vehicle position and orientation. The noise distribution of the multiple sensor readings is identified via a set of fuzzy logic (FL) controllers also used to update the measurement covariance matrix of the EKF. Artificial landmarks are recognized by the stereo vision system and distances between the vehicle and the landmarks are computed by both the laser range finder and the stereo vision system. Each FL controller is dedicated to one sensor and its primary function is to adjust the parameters of the sensor readings noise distribution. Range information, odometer measurements and FL controller outputs are inputs to the EKF that estimates the current position of the vehicle. As a case study, experiments with a skid steering mobile robot navigating indoors and outdoors are performed, and obtained experimental results demonstrate that the fuzzy EKF performs better than the EKF in terms of position accuracy.