ATI-CTLO:Adaptive Temporal Interval-based Continuous-Time LiDAR-Only Odometry

TL;DR

This letter proposes an Adaptive Temporal Interval-based Continuous-Time LiDAR-only Odometry (ATI-CTLO), which is based on straightforward and efficient linear interpolation and can flexibly adjust the temporal intervals between control nodes according to the motion dynamics and environmental degeneracy.

Abstract

The motion distortion in LiDAR scans caused by aggressive robot motion and varying terrain features significantly impacts the positioning and mapping performance of 3D LiDAR odometry. Existing distortion correction solutions often struggle to balance computational complexity and accuracy. In this work, we propose an Adaptive Temporal Interval-based Continuous-Time LiDAR-only Odometry, utilizing straightforward and efficient linear interpolation. Our method flexibly adjusts the temporal intervals between control nodes according to the dynamics of motion and environmental characteristics. This adaptability enhances performance across various motion states and improves robustness in challenging, feature-sparse environments. We validate the effectiveness of our method on multiple datasets across different platforms, achieving accuracy comparable to state-of-the-art LiDAR-only odometry methods. Notably, in scenarios involving aggressive motion and sparse features, our method outperforms existing solutions.

Figures (8)

• Figure 1: Motion distortion correction. (a): A single frame of the point cloud is distorted due to aggressive turning. (b): Correction distortion under constant velocity assumption (The mapping result of KISS-ICP vizzo2023kiss). (c): Our method, utilizing dynamic temporal interval control, ensures more accurate mapping.
• Figure 2: System Overview. Firstly, LiDAR scans segmentation is based on PCA pre-assessment, followed by solving through a sliding window approach (window's size $M$: 3). Simultaneously, degeneracy management merges cloud segments when degeneracy occurs.
• Figure 3: PCA Pre-assessment. A sudden change in the eigenvectors $\boldsymbol{v}_i$ indicates an aggressive turn in motion. In the upper image, the nodes with numbers show significant changes in the PCA direction. However, only the yellow nodes (2, 3, 4) are due to motion turns, while the red nodes (1, 5) result from environmental changes, such as the robot moving between a room and a hallway. The line chart below records the maximum PCA direction change $\boldsymbol{\varphi}_{max}$ (blue) and the maximum eigenvalue change $\mathcal{V}_{max}$ (yellow) throughout the entire trajectory. At nodes 1 and 5, there is a significant change in $\mathcal{V}_{max}$, and we filter out erroneous estimates (red nodes) by setting a threshold (red line).
• Figure 4: Degeneracy Detection: Each column $\mathbf{I}_j$ of the contribution matrix $\mathbf{I}$ represents the projection of environmental information in the corresponding principal direction. In the contribution evaluation, we calculate the contribution of each column $\mathbf{I}_j$ and use $l_c$ and $l_s$ to count the number of information elements $\mathbf{I}_{(i,j)}$ with varying levels of contribution. Then different thresholds $\kappa_1, \kappa_2, \kappa_3$ are used to classify the contributions into various levels, eventually the localizability of each direction is divided into three levels: {localizable, partially localizable, nonlocalizable}. Degeneracy Management: We enhance the number of effective information elements by increasing the sampling rate (decreasing $D_s$) for partially localizable cases. In nonlocalizable cases, subsequent cloud segment $\mathbf{S}_{i+3}$ is merged to gather more valid information.
• Figure 5: Results of LiDAR odometry on street_07 sequence of M2DGR. (a) Our method, the white box marks the position where CT-ICP drifts. (b) CT-ICP shows significant errors at the marked position in (a) . (c) Trajectory results aligned with the ground truth. (d) A zoomed-in view, where orange boxes indicate increased temporal intervals between control nodes due to degeneracy management, and black boxes indicate reduced intervals due to PCA-assessed motion state changes.