Autonomous mobile robots require centimeter-level localization accuracy at key sites (e.g., docking stations, loading zones, and charging points) across diverse environments such as manufacturing, logistics, and healthcare. Conventional SLAM-based techniques typically provide sufficient performance for global navigation; however, they fall short in achieving the precise localization needed near target points. Auxiliary methods using QR codes or fiducial markers enhance localization precision, yet they compromise flexibility and scalability in industrial contexts owing to significant installation and maintenance costs. Therefore, this study proposes a novel hybrid framework that integrates a multi-resolution map representation with a region-segmented localization architecture to address these challenges. The framework partitions the target vicinity into three regions—delivery area, hysteresis zone, and docking zone—and employs a modular algorithm tailored to the precision demands of each region. This approach ensures real-time precise localization without relying on physical markers and enables flexible adaptation to modifications at the target point. Specifically, the outcomes of Monte Carlo localization, a probabilistic global localization method, serve as initial estimates for iterative closest point registration, thereby ensuring both global robustness and local precision. The proposed framework was validated through simulations and experiments using real industrial robots, demonstrating enhanced accuracy and robustness relative to conventional 2D SLAM methods, reducing the mean docking error in simulation from 7.97–12.56 cm to 4.48–6.77 cm, and suppressing large-error cases in real-world experiments by lowering the 90th-percentile docking error from 67.4–88.7 cm to 3.02–17 cm.
Moon et al. (Thu,) studied this question.