Reproduce “D$^2$-LIO: Enhanced Optimization for LiDAR-IMU Odometry Considering Directional Degeneracy” on OpenResearch

About this paper

Guodong Yao, Hao Wang, Qing Chang

Researchers at Beihang University developed D²-LIO, an enhanced LiDAR-IMU odometry framework that improves localization accuracy and robustness in environments with geometric degeneracy. This approach achieves lower absolute trajectory error across diverse challenging datasets, for instance, reducing ATE to 0.10m in a tunnel scenario compared to 0.13-0.22m for other methods, by incorporating an adaptive per-point outlier rejection and a degeneracy-aware pose estimation strategy.

Key insights

• A per-point adaptive outlier rejection strategy, accounting for ego-motion and individual point range, significantly improves the robustness of feature matching by consistently managing spatially variant distortions.
• Directly incorporating degeneracy awareness into the multi-sensor fusion optimization, through adaptive weighting informed by geometric observability and IMU uncertainty, provides a principled approach to mitigate ill-conditioned problems.
• Analyzing the eigenvalues of the Hessian matrix allows for quantitative assessment of degeneracy and enables adaptive regularization, which enhances numerical stability in previously ill-constrained directions.

Summary by alphaXiv.