Session D4: Robust Navigation Using Alternative Navigation Sensors and Solutions

D² LIO: A Tightly-Coupled LiDAR-Inertial Odometry with Degeneracy Detection
Daifang Huang, Yong Li, Zhihang Qu, and Wenhui Yang, University of Electronics Science and Technology of China
Location: Johnson (First Floor)

Peer Reviewed

LiDAR , as an active perception sensor, has shown great potential in robotic applications. However, in degenerate environments such as tunnels and underground mines, LiDAR-based SLAM systems often suffer from significant performance degeneracy due to sparse structural features and the absence of GNSS signals. Existing approaches typically rely on multi-sensor fusion for pose correction without explicitly identifying the degenerate directions, which may compromise system accuracy and robustness. To address this, we propose a robust and lightweight LiDAR-Inertial Odometry (LIO) system that incorporates a novel online LiDAR degeneracy detection module. The degeneracy mechanism of LiDAR is first analyzed from the perspective of error perturbation, and a quantitative metric—referred to as the LiDAR Degeneracy Factor—is introduced to assess the observability of state constraints. Based on this metric, the proposed detection module identifies degenerate directions without relying on empirical thresholds and explicitly accounts for the coupling between rotational and translational components in the state space. In addition, a lightweight remapping strategy is designed to suppress the influence of degenerate directions while reinforcing the credibility of valid kinematic information, thereby improving the stability and precision of state estimation under degenerate conditions. This design enables the system to operate effectively in GNSS-denied environments and under sensor-degrading circumstances without the aid of visual inputs. Extensive experiments demonstrate that the proposed method achieves superior accuracy and robustness compared to state-of-the-art LIO systems, while maintaining low computational overhead. The entire framework is modular, generalizable, and easily integrable into existing LIO pipelines, making it a practical and reliable solution for autonomous navigation in challenging real-world environments.