Enhanced LiDAR SLAM Backend with Multi-Session Mode for Long-Term Association

Intelligent devices necessitate long-term, precise pose estimation and mapping to effectively perceive and describe the real world. To address this challenge, we propose an enhanced optimization and mapping module for the long-term association LiDAR SLAM back-end, based on a multi-session mode. We utilize an iVox-based point cloud registration method to enhance the speed of loop detection and dynamic point removal in point cloud maps. By employing a Distributed Pose Graph Optimization (DPGO) method, we associate data from multiple LiDAR sessions, facilitating information sharing between pose graphs and reducing their complexity. This approach significantly improves the speed of global pose optimization. Experimental results regarding computational efficiency, trajectory accuracy, and the quality of long-term point cloud maps demonstrate that our module enhances the existing LT -Mapper algorithm. Our module effectively reduces the time required for point cloud registration, enhances both the efficiency and accuracy of pose map optimization, and results in more precise long-term pose estimation.