GenZ-ICP: Generalizable and Degeneracy-Robust LiDAR Odometry Using an Adaptive Weighting

TL;DR

This study revisited both point-to-plane and point-to-point error metrics and proposed a method that leverages their strengths in a complementary manner called GenZ-ICP, which exhibits high adaptability to various environments and resilience to optimization degradation in corridor-like degenerative scenarios by preventing ill-posed problems during the optimization process.

Abstract

Light detection and ranging (LiDAR)-based odometry has been widely utilized for pose estimation due to its use of high-accuracy range measurements and immunity to ambient light conditions. However, the performance of LiDAR odometry varies depending on the environment and deteriorates in degenerative environments such as long corridors. This issue stems from the dependence on a single error metric, which has different strengths and weaknesses depending on the geometrical characteristics of the surroundings. To address these problems, this study proposes a novel iterative closest point (ICP) method called GenZ-ICP. We revisited both point-to-plane and point-to-point error metrics and propose a method that leverages their strengths in a complementary manner. Moreover, adaptability to diverse environments was enhanced by utilizing an adaptive weight that is adjusted based on the geometrical characteristics of the surroundings. As demonstrated in our experimental evaluation, the proposed GenZ-ICP exhibits high adaptability to various environments and resilience to optimization degradation in corridor-like degenerative scenarios by preventing ill-posed problems during the optimization process.

Figures (5)

• Figure 1: LiDAR mapping result of GenZ-ICP in the Long_Corridor sequence of the SubT-MRS dataset zhao2024cvpr. The accumulated map is shown in gray, while the current scan, classified by our proposed adaptive weighting, is colored in light blue where the point-to-plane error metric is applied for planar regions and in red where the point-to-point error metric is applied for non-planar regions. Note that all zoomed-in images are from the same scan, and the visualized coordinate corresponds to the robot’s body frame. Our GenZ-ICP adaptively utilizes both error metrics by reflecting the geometrical characteristics of the surroundings, achieving robustness across various environments, particularly in corridor-like environments. The camera image is included only for better understanding of the scene.
• Figure 2: Flowchart of the proposed system. The current scan from the LiDAR frame is transformed into the map frame using previous odometry and enters the ICP loop with the local map. Next, GenZ-ICP, which applies an adaptive weighting scheme, robustly estimates the pose in various environments, especially in degenerative scenarios.
• Figure 3: (a)-(b) Qualitative results before and after applying adaptive weighting, showing the adaptability to the changing geometrical characteristics of the surroundings in the short experiment sequence of Newer College dataset ramezani2020iros. $\medcirc$ denotes the start and end points of the sequence, and $\times$ indicates the divergence point. (a) Without adaptive weighting, i.e., $\alpha=0.5$, odometry diverged due to its inability to reflect the geometrical characteristics of the surroundings. (b) A color map with the adaptive weight $\alpha$ represents each pose. More structured surroundings result in $\alpha$ values closer to one. Thus, the narrow corridor without windows, highlighted as a zoomed box, has a higher $\alpha$ value than other scenes.
• Figure 4: Qualitative results for the Long_Corridor sequence of SubT-MRS dataset zhao2024cvpr. $\medcirc$ and $\square$ denote start and end points, respectively. (a) is the ground truth map. In (b), the system using point-to-point besl1992tpami error metric exhibited reduced pose drift but resulted in an inaccurate map. Moreover, in (c), (d), and (e), the systems using point-to-plane rusinkiewicz2001IntConfThreeDDigitalImagingAndModeling or G-ICP segal2009rss-based error metric exhibited pose drift due to degeneracy, respectively, resulting in significantly different arrival positions. However, in (f), our approach was robust against degeneracy and showed the most accurate result.
• Figure 5: Box plot of condition number in corridor scenarios. Our method demonstrated the lowest condition number across all sequences. A lower condition number indicates that the numerical condition of a system is stable cheney1998numerical. Our method prevents mathematically ill-posed problems in the optimization process, resulting in resilience to optimization degradation in corridor-like degenerative scenarios. The $\text{****}$ annotations indicate measurements with $p$-value $<10^{-4}$ after a paired $t$-Test.