Data association using relative compatibility of multiple observations for EKF-SLAM

Abstract

Correct data association is crucial to perform self-localization and map building for mobile robot. The nearest neighbor method based on the maximum likelihood is widely used. However, this algorithm has two problems, possibility of false association and spurious association. These problems happen more severely when the vehicle pose error is large and the covariance does not represent the uncertainty correctly. In this paper, a data association method which applies the concept of pairwise relative compatibility to the probabilistic data association problem is proposed. The proposed method handles the false and spurious association problems effectively. We prove its performance by the EKF-SLAM simulations and experiments and the results show that the proposed data association provides reliable data association.