DB-TSDF fuses LiDAR point clouds into a dense signed distance field using fixed-size directional kernels. The per-frame update cost is bounded and largely independent of map resolution, enabling high-resolution CPU mapping.
DB-TSDF Directional Bitmask-based TSDF for Efficient Volumetric Mapping
CPU-only TSDF mapping with constant per-scan time via directional kernels and bitmask distance encoding. Open-source implementation available.
Overview
Method
New: A 16-bit version is now available, offering improved memory efficiency and performance.
Each LiDAR return selects a precomputed, direction-dependent kernel (aligned to azimuth/elevation bins) and applies a bitwise AND update to the voxel’s 32-bit distance mask. A hemispherical shadow region increments a saturating counter to confirm occupancy. The compact 8-byte voxel layout is cache-friendly and parallelizable.
Results
Qualitative results and timings. See the paper for full quantitative comparisons.
Citation
If you use DB-TSDF in your research, please cite our ICRA 2026 paper:
BibTeX
@inproceedings{maese2026dbtsdf,
title={DB-TSDF: Directional Bitmask-based Truncated Signed Distance Fields for Efficient Volumetric Mapping},
author={Maese, Jose Enrique and Caballero, Fernando and Merino, Luis},
booktitle={2026 IEEE International Conference on Robotics and Automation (ICRA)},
year={2026}
}
Funding
This work was supported by PICRA 4.0 (PLEC2023-010353) funded by MCIN/AEI and the Spanish Research Agency; and INSERTION (PID2021-127648OB-C31), funded by the "Agencia Estatal de Investigación - Ministerio de Ciencia, Innovación y Universidades" and the "European Union NextGenerationEU/PRTR".
