A Real-Time Localization Method Using LiDAR Beacons

In GNSS-denied environments, fusing LiDAR scan with prior 3D maps has proven to be an effective approach for global localization. This study introduces a real-time global localization framework using LiDAR beacons, which innovatively integrates the strengths of Monte Carlo Localization (MCL) and match-based methods. The proposed system features a two-stage process: first, a hybrid place recognition that combines rapid match-based initialization and robust MCL is designed based on LiDAR beacons during the localization initial phase. Then, an adaptive environment error model of LiDAR odometry is developed to establish spatial distance constraints, significantly enhancing both localization accuracy and robustness. Comprehensive experiments demonstrate that the proposed method reduces the root mean square error by $\mathbf{2 0 \%} \mathbf{- 5 0 \%}$ compared to existing approaches across multiple datasets, including KITTI and NCLT. Additionally, the initial convergence distance is reduced to 1/10 of traditional MCL, effectively addressing challenges such as robot “wake-up” and “kidnapping” in complex scenarios.