Dynamic Noise Adaptation in the Motion Model of Monte Carlo Localization for Consistent Localization

Precise position estimation is essential for mobile robots to operate autonomously. In industrial environments that require precision tasks such as docking—including structured indoor facilities such as hospitals, factories, and warehouses—highly accurate localization is often necessary, with accuracy demands ranging from the centimeter to millimeter level depending on the application. Various registration-based localization algorithms have been investigated in response to this requirement. However, fundamental limitations exist, such as a high dependency on initial position estimates, increased computational load, and difficulties in ensuring real-time performance in large-scale environments. The proposed method introduces a dynamic noise adaptation (DNA) technique applicable to the Monte Carlo localization (MCL) algorithm, a particle filter-based localization method, to overcome these limitations. The proposed algorithm improves real-time localization accuracy and estimation consistency by dynamically optimizing the motion noise of MCL using the non-penetration rate, which can serve as a reliability metric in light detection and ranging (LiDAR)-based localization. The proposed algorithm was evaluated in comparison with the expansion Monte Carlo localization 2 (EMCL2) algorithm in both simulation and real-world environments. In the simulated environment, the proposed method achieved lower localization error with respect to the ground truth compared to EMCL2 and the improved adaptive Monte Carlo localization (AMCL) method incorporating a virtual motion model. In real-world experiments, localization performance was evaluated through comparison with a reference trajectory, and the proposed algorithm consistently demonstrated reduced localization error.