Real-Time Submap Trimming-Based Map Updating for Long-Term Localization of Mobile Robot in Dynamic Environments

Effective map updating is crucial in maintaining accurate long-term localization for mobile robots. However, existing studies encounter difficulties in real-time updates of the global map for dynamic environments and are prone to overwhelming amounts of redundant information, which can easily cause an information explosion. To address these challenges, this article presents a novel map-updating method using real-time submap trimming (RTST). First, the proposed approach preliminarily selects redundant submaps with high overlap rates based on the centroid features of submaps in real time, and the observation gain model is further employed to observe the effectiveness of the submaps, thereby filtering disabled submaps from long-lived old submaps. Second, the filtered redundant submaps are removed from the global map to ensure that the map is always up-to-date online, avoiding information explosion, and ultimately improving long-term localization accuracy. Finally, to verify the effectiveness of the proposed method, we specially conducted long-term localization experiments in a real scenario using a low-computational capability mobile robot based on an embedded system. Our experimental results demonstrate that the proposed method significantly improves the mobile robot's real-time performance for map updates while ensuring long-term localization accuracy.