Robust localization in degenerate environments using uncertainty-aware LiDAR-inertial odometry

This paper presents an uncertainty-aware LiDAR-inertial odometry framework explicitly designed to enhance localization performance in geometrically challenging environments. Accurate and robust odometry systems, particularly for GPS-denied scenarios, demand precise uncertainty modeling, efficient mapping techniques, and effective management of sensor noise and degeneracies. Existing methods often struggle in structurally ambiguous or feature-poor contexts, such as tunnels and corridors. To address this, we introduce comprehensive uncertainty modeling at both sensor measurement and map representation levels, employing adaptive filtering and robust point-to-plane residual analysis. Our approach significantly improves localization consistency by dynamically weighting sensor inputs based on their estimated reliability. Experimental evaluations conducted on diverse datasets demonstrate superior robustness and generalizability of the proposed method, showcasing its potential for applications in underground navigation, industrial inspections, and disaster response, where traditional localization approaches frequently falter.