VoxLIO-BA: A LiDAR-Inertial Odometry with the Adjustment of Voxel Map and Historical Scan via Bundle Adjustment

Reducing the error drift of point cloud map is essential for Simultaneous Localization and Mapping (SLAM) systems. In this paper, a novel LiDAR-inertial SLAM system equipped with the ca-pability to Adjust Voxel Map and Historical Scan State, VoxLIO-BA, is proposed. This method maintains the map through a voxel-based structure that ensures efficient spatial hashing for map maintenance, en-abling precise scan-to-map registration in the odometry module while incorporating a sliding-window Bun-dle Adjustment(BA) module to jointly optimize the map and multi-scan poses. This dual optimization reduces cumulative map errors and enhances pose estimation accuracy. By feeding back adjusted poses to the odometry module, the system ensures map-odometry consistency while enhancing the reliability of odometric propagation. The evaluations on differ-ent public datasets demonstrate the system's superior performance in localization accuracy.