Autonomous robots localization in repetitive environments

Abstract

With the rapid development of autonomous robots these years, the localization and positioning algorithms are expected to be more and more accurate and robust. Currently, the most commonly used global localization method is the Global Navigation Satellite System (GNSS). However, when the application environment involves indoor or blocking scenarios, the GNSS signal can be unstable or even unable to acquire. In such cases, local localization methods using onboard sensors like LiDAR and cameras are widely used. However, these methods cannot work well in extremely similar and repetitive environments because the information captured by the sensors can be highly similar or identical at different locations. The onboard LiDAR/vision-based localization or simultaneous localization and mapping (SLAM) methods tend to wrongly localize the autonomous robots or even fail in such challenging scenarios. In recent years, ambient magnetic field (MF)-based localization has proven to be a feasible way to achieve meter-level accuracy in semi-indoor or indoor environments. Therefore, this project will explore the application of magnetic field information to solve the localization problem in highly repetitive environments. To solve the autonomous robots localization problem, many MF-based localization algorithms emerged. However, the existing methods have certain limitations, such as low-accuracy and high computational cost. The positioning potential of the ambient MF has not been fully investigated and utilized. Therefore, this report will explore and propose a novel global localization system with the hierarchical fusion of LiDAR-based and MF-based localization. The proposed method is compared with state-of-the-art MF-based localization methods, such as the particle filter (PF)-based algorithm and dynamic time warping (DTW)-based algorithm. The extensive experimental results demonstrate the advantageous robustness and accuracy of the proposed method.