Performance comparisons of particle-based and similarity measure-based kidnapping detectors in Monte Carlo localization

Abstract

This paper investigates new alternative approaches to detect the kidnapped robot problem event in Monte Carlo Localization. The approach is designed such that it can provide accurate detection in wide range kidnapping points and does not depend on the accuracy of localization. The underlying idea is based on the similarity measures of the environment seen by the robot at two consecutive time instances. Six different similarity measures are investigated and tested against particles weight-based detectors to see how good each detector’s ability to distinguish normal condition from kidnapping event, i.e. Discrimination Performance, under two different kidnapping scenarios. These simulations show that Two-Dimensional Dynamic Time Warping promises better general accuracy across all kidnapping points compared to particles-based detectors and other detectors based on shape similarity measure. The Consistency Performance also shows that it can maintain the accuracy even when the localization process is heavily disturbed.