Extended Kalman filtering for fuzzy modelling and multi-sensor fusion

TLDR

The study proposes using Extended Kalman Filtering to both extract fuzzy models from data and localize autonomous vehicles. EKF is applied to fuzzy model extraction and vehicle state estimation, outperforming the Gauss–Newton nonlinear least‑squares method and fusing odometric and sonar data for localization. EKF converges faster than Gauss–Newton, provides a convergence analysis, and achieves satisfactory localization accuracy in simulations.

Abstract

Extended Kalman Filtering (EKF) is proposed for: (i) the extraction of a fuzzy model from numerical data; and (ii) the localization of an autonomous vehicle. In the first case, the EKF algorithm is compared to the Gauss–Newton nonlinear least-squares method and is shown to be faster. An analysis of the EKF convergence is given. In the second case, the EKF algorithm estimates the state vector of the autonomous vehicle by fusing data coming from odometric sensors and sonars. Simulation tests show that the accuracy of the EKF-based vehicle localization is satisfactory.