← Back ICRA 2026

MAD-BA: 3D LiDAR Bundle Adjustment -- from Uncertainty Modelling to Structure Optimization

Krzysztof Cwian, Luca Di Giammarino, Simone Ferrari, Thomas Alessandro Ciarfuglia, Giorgio Grisetti, Piotr Skrzypczynski

PDF

AI summary

Key figure (auto-extracted from paper)

MAD-BA jointly optimizes LiDAR sensor poses and surfel-based 3D maps using a generalized uncertainty model, significantly improving mapping consistency and accuracy over existing methods.

LiDAR SLAM Bundle Adjustment 3D Mapping Uncertainty Modeling Surfel Optimization Pose Refinement

Problem

Current LiDAR SLAM systems typically optimize sensor trajectories and 3D structure separately, causing drift, noise accumulation, and inconsistent maps. This decoupling prevents globally consistent refinement in challenging or sparse environments.

Approach

The framework jointly refines poses and a surfel-based map representation by weighting measurements with a generalized LiDAR uncertainty model derived from beam divergence and incidence angles. It leverages kd-tree data association and Levenberg-Marquardt optimization to minimize geometric errors globally.

Key results

Jointly optimizes LiDAR poses and surfel-based structure for global consistency
Introduces a generalized, sample-based LiDAR uncertainty model for measurement weighting
Achieves lower trajectory and geometric errors than BALM2 and HBA across NC, VBR, and KITTI datasets
Provides open-source implementation for community reproducibility

Why it matters

Enables more accurate and robust 3D mapping for autonomous navigation, robotics, and AR/VR by resolving the long-standing decoupling of pose and structure optimization in LiDAR systems.

Abstract

The joint optimization of sensor poses and 3D struc- ture is fundamental for state estimation in robotics and related fields. Current LiDAR systems often prioritize pose optimization, with structure refinement either omitted or treated separately using implicit representations. This paper introduces a frame- work for simultaneous optimization of sensor poses and 3D map, represented as surfels. A generalized LiDAR uncertainty model is proposed to address less reliable measurements in varying scenarios. Experimental results on public datasets demonstrate improved performance over most comparable state-of-the-art methods. The system is provided as open-source software to support further research.

Index terms

Mapping Range Sensing SLAM