Improving Monte Carlo Localization Performance Using Strategic Navigation Policies

Abstract

An important problem in robotics is to determine and maintain the position of a robot that moves through a previously known environment with indistinguishable reference points. This problem is made difficult due to the inherent noise in robot movement and identification of reference points and due to multiple identical reference points. Monte Carlo Localization (MCL) is a frequently used technique to solve this problem and its performance intuitively depends on how the robot explores the map. In this paper, we evaluate the performance of MCL under different navigation policies. In particular, we propose a novel navigation policy that aims in reducing the uncertainty in the robot's location by making a greedy movement at every step. We show that this navigation policy can significantly outperform random movements, particularly when the map has few reference points. Moreover, differently from random movements, the performance of the proposed navigation policy is not monotonic with the number of reference points on the map.