A Map Construction and Maintenance Framework for Long-term Navigation Based on LiDAR

We propose a method for constructing and maintaining a navigation map in dynamic environments using laser scan information. During long-term robot operation, changes in environmental information are unavoidable. Accurate map construction and real-time map updates are crucial for ensuring reliable autonomous navigation over an extended period. Moreover, the dynamic environmental information in the global map can adversely affect the accuracy of LiDAR-based simultaneous localization and mapping algorithms. Given the sparsity of real-time laser scan data in comparison to previous maps, this paper presents a map construction algorithm and a map update algorithm that utilize obstacle-level segmentation. These algorithms demonstrate improved noise filtering capabilities and dynamic information detection, thereby ensuring the accuracy and reliability of the navigation map. We evaluated our approach through experiments conducted in both virtual simulation and real-world physical environments. The results demonstrate its effectiveness in enhancing map construction and map update performance, as well as providing reliable map information for autonomous robot navigation.