Fusion of Odometry with Magnetic Sensors Using Kalman Filters and Augmented System Models for Mobile Robot Navigation

Abstract

This paper presents a comparative study of two data fusion methods for high precision mobile robot's pose estimation. Odometric data, provided by wheels encoders, are fused with data from magnetic markers detection. One of the methods uses an extended Kalman filter and the other uses a linear Kalman filter whose application is made possible by using an augmented state system model. The measurement system is composed by wheel encoders and two magnetic sensing rulers, one on the front and the other on the rear of the mobile robot, for magnetic markers detection. Simulation results with a very realistic approach are presented. 1 I. INTRODUCTION Odometry is a method of localization for land vehicles used with generality, based on wheels encoder's data. This method however, based on dead reckoning, produces cumulative errors which in case of no compensation, result over time in an appreciated pose deviation. One possible way of errors compensation is the use of fusion of odometry with other sensorial data which makes possible its correction. The use of magnetic markers embedded in the ground, may supply the kind of necessary information for the odometric data calibration, in a way, to have available a pose estimation of the vehicle with the required precision for autonomous navigation. Sensor fusion regarding mobile robot localization should be turning to computing algorithms which produce its optimal estimation. The Kalman filter and its variants, verify this requirement being for this reason largely used. The odometry and measurement models, involved in pose estimation by using landmarks such as magnetic markers, are nonlinear, and therefore the use of an extended Kalman filter (EKF) would become inevitable in the fusion process. However, by using an augmented state a linear Kalman filter (LKF) can be applied instead of the EKF. This paper presents a comparative study of two data fusion methods for high precision vehicles pose estimation. One based on an