Detection strategy for kidnapped robot problem in landmark-based map Monte Carlo Localization

Abstract

This paper proposes a new method to detect the kidnapped robot problem event in Monte Carlo Localization. The method is designed such that it can provide accurate detection in wide range of particles' convergence level and does not depend too much on the re-localization/recovery process. The proposed method combines the difference in particle's weight, maximum current weight, and difference in particles' standard deviation. The addition of these two parameters is believed to be superior to a pure maximum current weight parameter for kidnapping detection. A series of simulation tests are executed to prove it. These simulations show that the proposed method outperforms the maximum current weight parameter in terms of accuracy, ability to detect kidnapping during early stage of localization, and independency towards the success of the re-localization process.