Papers
Topics
Authors
Recent
Gemini 2.5 Flash
Gemini 2.5 Flash
102 tokens/sec
GPT-4o
59 tokens/sec
Gemini 2.5 Pro Pro
43 tokens/sec
o3 Pro
6 tokens/sec
GPT-4.1 Pro
50 tokens/sec
DeepSeek R1 via Azure Pro
28 tokens/sec
2000 character limit reached

SLAMesh: Real-time LiDAR Simultaneous Localization and Meshing (2303.05252v1)

Published 9 Mar 2023 in cs.RO

Abstract: Most current LiDAR simultaneous localization and mapping (SLAM) systems build maps in point clouds, which are sparse when zoomed in, even though they seem dense to human eyes. Dense maps are essential for robotic applications, such as map-based navigation. Due to the low memory cost, mesh has become an attractive dense model for mapping in recent years. However, existing methods usually produce mesh maps by using an offline post-processing step to generate mesh maps. This two-step pipeline does not allow these methods to use the built mesh maps online and to enable localization and meshing to benefit each other. To solve this problem, we propose the first CPU-only real-time LiDAR SLAM system that can simultaneously build a mesh map and perform localization against the mesh map. A novel and direct meshing strategy with Gaussian process reconstruction realizes the fast building, registration, and updating of mesh maps. We perform experiments on several public datasets. The results show that our SLAM system can run at around $40$Hz. The localization and meshing accuracy also outperforms the state-of-the-art methods, including the TSDF map and Poisson reconstruction. Our code and video demos are available at: https://github.com/lab-sun/SLAMesh.

Real-time LiDAR SLAM System with Simultaneous Localization and Meshing

The paper "SLAMesh: Real-time LiDAR Simultaneous Localization and Meshing" presents a novel approach to LiDAR-based simultaneous localization and mapping (SLAM). This system is uniquely designed to integrate real-time mesh mapping with localization, aimed at addressing limitations of current point-cloud based SLAM systems.

Introduction and Motivation

Current LiDAR SLAM systems commonly employ point-cloud maps to represent 3D environments. These maps, despite seeming dense, are sparse upon closer inspection, requiring additional processing for applications like navigation. Mesh models, due to lower memory costs and dense mapping capabilities, have gained interest. Traditional approaches create mesh maps offline, separating meshing from localization, and failing to leverage the benefits that simultaneous processes could offer. This paper introduces the first CPU-only real-time LiDAR SLAM system that builds and utilizes mesh maps on-the-fly, integrating both localization and meshing.

Methodology

The SLAMesh system is distinguished by its use of a Gaussian process (GP) reconstruction for real-time mesh map development. This approach enables efficient building, registration, and updating of mesh maps directly in a continuous process. The system operates on several public datasets, demonstrating the capacity of SLAMesh to run at approximately 40Hz with superior localization and meshing accuracy.

Meshing Strategy

The approach tackles the inherent complexity of mesh updating through a reconstruction and connection method. Vertices in the mesh are distributed uniformly through a local surface reconstruction via GP, enabling the establishment of a mesh map in real-time. This approach ensures fast data querying and reduces the inaccuracies associated with discretization.

Point-to-Mesh Registration

SLAMesh employs a point-to-mesh registration method, capitalizing on mesh face normals to enhance localization accuracy. This method establishes correspondences based on vertex locations, facilitating a robust registration process through effective utilization of normal information from mesh faces.

Mesh Management

Efficient mesh management is key to SLAMesh’s real-time performance. The system limits its updates to 1-D predictions, thereby maintaining efficient data structures. By structuring data into independent cells, SLAMesh further reduces update complexities, benefiting from parallel processing through multithreading.

Results

The experimental evaluation establishes SLAMesh’s superiority across various metrics. On public datasets, the system demonstrates high-quality mesh maps and lower computational overhead compared to existing methods. These results are supported by memory efficiency and processing time, validating SLAMesh's applicability for large-scale environments.

The numerical results indicate SLAMesh's average odometry errors at 0.676% in translation and 0.291 deg/100m in rotation, outperforming traditional point-cloud, NDT, and surfel-based methods. The mesh maps created by SLAMesh preserve detailed structures, achieving better precision and recall metrics, establishing their utility in both robotics and mapping applications.

Implications and Future Developments

The integration of real-time meshing with localization paves the way for SLAM systems that are more efficient and accurate, potentially transforming autonomous navigation and 3D reconstruction tasks. The methodology offers scalable solutions with reduced memory requirements, which are crucial for resource-constrained robotic platforms.

Future research may explore extending SLAMesh capabilities to various sensor modalities, enhancing robustness in dynamic environments, and further reducing computational loads through optimized parallel processing schemes. The open-source nature of the SLAMesh system also invites contributions, fostering ongoing innovations in real-time SLAM and meshing applications.

User Edit Pencil Streamline Icon: https://streamlinehq.com
Authors (4)
  1. Jianyuan Ruan (2 papers)
  2. Bo Li (1107 papers)
  3. Yibo Wang (111 papers)
  4. Yuxiang Sun (39 papers)
Citations (18)
Github Logo Streamline Icon: https://streamlinehq.com