An Efficient and Robust Approach to Solve the Kidnapped Robot Problem Considering Time Variation

Abstract

This paper proposes an efficient and robust localization recovery method considering time variation for Monte Carlo localization (MCL), which can handle the kidnapped robot problem (KRP) with improved recovery speed and success rate. The presence of particles near the real robot’s pose is necessary for the localization recovery of MCL. Therefore, the generation number and position of random particles are the vital factors to solve KRP. It is generally assumed that the robot cannot move instantaneously, so the size of the search space where robot may appear should be time-dependent after the KRP occurs. Given these considerations, a restricted search space is firstly constructed as a set subject to a time-varying normal distribution, which can effectively narrow the search space and estimate the probability that the robot may appear. In addition, a short-long term random particle generation strategy considering time variation is designed to availably determine the number of random particles according to the change of the likelihood and time. And then, the random particles are spread into the restricted search space. Finally, the above particle set is integrated into the MCL for localization recovery. The effectiveness of the proposed method is verified through real scene experiments.