← Back ICRA 2026

DB-TSDF: Directional Bitmask-based Truncated Signed Distance Fields for Efficient Volumetric Mapping

Jose Enrique Maese, Luis Merino, Fernando Caballero

PDF

AI summary

Key figure (auto-extracted from paper)

DB-TSDF enables high-resolution, real-time 3D volumetric mapping on CPU-only platforms by using directional bitmask kernels, achieving accuracy comparable to GPU-accelerated methods without sacrificing speed.

Volumetric mapping CPU-only processing Truncated Signed Distance Fields Bitmask encoding Real-time 3D reconstruction LiDAR mapping

Problem

Existing TSDF/ESDF mapping methods heavily rely on GPU acceleration or suffer from computational costs that scale unfavorably with map resolution and update rate on CPUs, hindering real-time high-resolution mapping on constrained platforms.

Approach

The method integrates LiDAR point clouds into a voxel grid using precomputed directional bitmask kernels and bitwise operations, ensuring constant per-scan processing time regardless of global grid resolution, all implemented entirely on CPU with multi-threading.

Key results

Constant per-scan computational cost independent of global grid resolution
Achieves state-of-the-art F-score of 96.6% on the Mai City dataset
Maintains stable ~150 ms per-frame latency across resolutions from 0.3 m to 0.05 m
Compact 8-byte voxel structure enabling high-resolution mapping with minimal memory overhead

Why it matters

It provides a practical, high-resolution volumetric mapping solution for CPU-constrained robotic systems, eliminating the need for GPU acceleration while maintaining real-time performance and geometric accuracy.

Abstract

This paper presents a high-efficiency, CPU-only volumetric mapping framework based on a Truncated Signed Distance Field (TSDF). The system incrementally fuses raw LiDAR point-cloud data into a voxel grid using a directional bitmask-based integration scheme, producing dense and consis- tent TSDF representations suitable for real-time 3D reconstruc- tion. A key feature of the approach is that the processing time per point-cloud remains constant, regardless of the voxel grid resolution, enabling high resolution mapping without sacrificing runtime performance. In contrast to most recent TSDF/ESDF methods that rely on GPU acceleration, our method operates entirely on CPU, achieving competitive results in speed. Ex- periments on real-world open datasets demonstrate that the generated maps attain accuracy on par with contemporary mapping techniques. The source code is publicly available at https://github.com/robotics-upo/DB-TSDF.

Index terms

Mapping