Adaptive-LIO: Enhancing Robustness and Precision through Environmental Adaptation in LiDAR Inertial Odometry

TL;DR

Adaptive-LIO tackles robustness and precision in LiDAR–Inertial Odometry under GPS-denied and challenging underground/urban settings by integrating three adaptive components: observability-driven adaptive frame segmentation to reduce motion distortion, IMU saturation detection with motion modality switching between LIO and LO, and a hash-based multi-resolution voxel map that preserves map density across distances. It combines continuous-time ICP deskewing with a loosely coupled fusion framework and a modular pipeline that updates poses and maps in real time. The approach demonstrates improved end-to-end translation error and reduced degeneracy across open-to-narrow transitions and extreme motions, while maintaining real-time performance on commodity hardware; the method is open-source on GitHub. These contributions are significant for robust localization and mapping in scenarios with hardware timing limitations or dynamic geometry, enabling more reliable autonomous operation in IoT-rich environments.

Abstract

The emerging Internet of Things (IoT) applications, such as driverless cars, have a growing demand for high-precision positioning and navigation. Nowadays, LiDAR inertial odometry becomes increasingly prevalent in robotics and autonomous driving. However, many current SLAM systems lack sufficient adaptability to various scenarios. Challenges include decreased point cloud accuracy with longer frame intervals under the constant velocity assumption, coupling of erroneous IMU information when IMU saturation occurs, and decreased localization accuracy due to the use of fixed-resolution maps during indoor-outdoor scene transitions. To address these issues, we propose a loosely coupled adaptive LiDAR-Inertial-Odometry named \textbf{Adaptive-LIO}, which incorporates adaptive segmentation to enhance mapping accuracy, adapts motion modality through IMU saturation and fault detection, and adjusts map resolution adaptively using multi-resolution voxel maps based on the distance from the LiDAR center. Our proposed method has been tested in various challenging scenarios, demonstrating the effectiveness of the improvements we introduce. The code is open-source on GitHub: \href{https://github.com/chengwei0427/adaptive_lio}{Adaptive-LIO}.

Figures (13)

• Figure 1: System pipeline.
• Figure 2: Frame segmentation strategy.
• Figure 3: LO and LIO switching.
• Figure 4: Multi-resolution voxel map. voxel $i$ contains a set of points $\mathbf{V}_i$ , and each point in $\mathbf{V}_i$ defines a nearest neighbor, where the relationship between points, voxels and maps is denoted as $\mathbf{p}_i \in \mathbf{V}_i\subset \mathbf{\mathcal{M}}_i, \mathbf{\mathcal{M}}_i \subset \mathcal{M}$.
• Figure 5: Nearest neighbor search. Finding a given point $\mathbf{p}_i$ of the neighboring voxels, and then search through the radius to get the nearest neighbors $\left\{\mathbf{q}_i^W\right\}$ of a given point $\mathbf{p}_i$. Here we set the number of nearest neighbor voxels to 6.