NV-LIO: LiDAR-Inertial Odometry using Normal Vectors Towards Robust SLAM in Multifloor Environments (2405.12563v2)

Abstract: Over the last few decades, numerous LiDAR-inertial odometry (LIO) algorithms have been developed, demonstrating satisfactory performance across diverse environments. Most of these algorithms have predominantly been validated in open outdoor environments, however they often encounter challenges in confined indoor settings. In such indoor environments, reliable point cloud registration becomes problematic due to the rapid changes in LiDAR scans and repetitive structural features like walls and stairs, particularly in multifloor buildings. In this paper, we present NV-LIO, a normal vector based LIO framework, designed for simultaneous localization and mapping (SLAM) in indoor environments with multifloor structures. Our approach extracts the normal vectors from the LiDAR scans and utilizes them for correspondence search to enhance the point cloud registration performance. To ensure robust registration, the distribution of the normal vector directions is analyzed, and situations of degeneracy are examined to adjust the matching uncertainty. Additionally, a viewpoint based loop closure module is implemented to avoid wrong correspondences that are blocked by the walls. The propsed method is validated through public datasets and our own dataset. To contribute to the community, the code will be made public on https://github.com/dhchung/nv_lio.

• The paper introduces NV-LIO, a normal vector-based LiDAR-Inertial Odometry framework that significantly enhances indoor SLAM performance.
• It employs a novel window-based normal vector extraction and matching algorithm to improve point cloud registration in confined environments.
• Experimental results on public and in-house datasets demonstrate reduced errors and robust loop closure despite complex multifloor structures.

NV-LIO: A Normal Vector-Enhanced LiDAR-Inertial Odometry Framework for Indoor SLAM

The paper under discussion introduces a novel normal vector-based tightly-coupled LiDAR-Inertial Odometry (LIO) framework, termed NV-LIO, designed to enhance simultaneous localization and mapping (SLAM) performance in challenging indoor environments, particularly those with multifloor structures. Traditional LIO algorithms often exhibit degraded performance in these settings due to rapid changes in LiDAR scans and repetitive structural features. The NV-LIO framework addresses these challenges by employing normal vectors for reliable point cloud registration, focusing on robustness and accuracy in confined spaces.

Methodological Innovations

The NV-LIO framework incorporates several innovative techniques to improve indoor SLAM performance:

1. Normal Vector Extraction: The framework extracts normal vectors from dense 3D LiDAR scans projected into range images. Using these vectors in the correspondence search enhances the accuracy of point cloud registration. This process includes a novel window-based approach to compute normal vectors, ensuring stability even in confined indoor environments with high-density point clouds.
2. Matching with Normal Vectors: Registration is performed by considering both the nearest neighbor and the angle difference of the normal vectors, resulting in more accurate correspondence search and alignment. This method critically reduces errors associated with traditional scan-to-scan matching in narrow spaces.
3. Degeneracy Detection and Handling: The paper introduces a degeneracy detection algorithm to identify and manage scenarios where the normal vector distribution indicates potential localization ambiguities, such as in long corridors or stairwells. By analyzing the distribution of normal vectors through principal component analysis, and adjusting the matching uncertainty accordingly, the framework maintains robust performance under such conditions.
4. Viewpoint-Based Loop Closure: A novel loop closure module based on viewpoint analysis is implemented to prevent incorrect correspondences in indoor environments with repetitive features. By projecting the normal cloud into a range image under the current viewpoint and excluding obstructed points, this module effectively mitigates errors typically encountered in multifloor transitions.

Experimental Validation

The methodological contributions of NV-LIO were validated through a series of experiments using both public datasets (SubT-MRS and Newer College datasets) and an in-house dataset based in various buildings on the KAIST campus. The results demonstrate the framework's effectiveness:

• SubT-MRS Dataset: The framework excelled in multifloor environments, especially in reliably mapping stairwells and narrow corridors, where other state-of-the-art methods like Faster-LIO and DLIO exhibited significant drift or incorrect keyframe matching.
• Newer College Dataset: NV-LIO showed competitive or superior performance in terms of root mean square error across various scenarios, managing rapid motion and very narrow environments better than other compared methods. Although computational costs were higher, they were justified by the enhanced accuracy and consistency.
• KAIST Campus Dataset: NV-LIO successfully performed online SLAM across diverse buildings with different structural complexities, such as sports complexes and research facilities, showcasing its adaptation to varying indoor characteristics.

Implications and Future Directions

The introduction of normal vector-based techniques in LIO frameworks paves the way for improved SLAM performance in constrained indoor environments. This work's implications extend to multiple robotic applications, including firefighting, security, and delivery robots, which operate in complex indoor settings. The enhanced accuracy and robustness provided by NV-LIO can significantly impact these domains, enabling more reliable and autonomous navigation.

Going forward, further research might explore optimized normal vector extraction methods to reduce computational overhead without compromising accuracy. Additionally, integrating machine learning techniques for adaptive parameter tuning during normal vector extraction and correspondence matching could enhance the framework's applicability to even more dynamic and complex environments.

In conclusion, NV-LIO represents a significant advancement in indoor SLAM by leveraging the power of normal vectors for robust LiDAR-inertial odometry. The framework's validated performance across various datasets highlights its potential as a critical tool for indoor navigation in multifloor structures, promising to enhance the capabilities of autonomous mobile robots.