The paper has been accepted by IEEE RAL 2026 and is available on IEEE Xplore Early Access.
The code is currently being organized and will be open-sourced soon.
R-VoxelMap is a voxel mapping method that improves localization accuracy in online LiDAR odometry by using a geometry-driven recursive plane fitting strategy. Our code is based on VoxelMap and primarily addresses the issues where VoxelMap and its variants typically fit and check planes using all points in a voxel, leading to parameter deviation due to outliers, over-segmentation of large planes, and incorrect merging across different physical planes. The main changes are as follows:
-
R-VoxelMap performs an outlier detect-and-reuse pipeline in each recursive iteration, effectively suppressing outlier influence, improving map accuracy, and reducing plane over-segmentation.
-
R-VoxelMap uses a point distribution-based plane validity check strategy, projecting and clustering point clouds on the RANSAC-fitted plane to prevent incorrect merging of different physical planes.
The framework of Lidar(-inertial) odometry based on R-VoxelMap.
The construction steps of R-VoxelMap. (outlier detect-and-reuse pipeline)
Our code is built on top of VoxelMap. We would like to acknowledge the contributions of the VoxelMap team for providing such a valuable framework.