Scholarlink

<p>Simultaneous Localisation and Mapping (SLAM) stands as a pivotal technology for large-scale positioning and map construction. During mapping operations in expansive environments, mobile robots incur inherent motion errors alongside point cloud overlaps and drift, consequently diminishing mapping accuracy. Laser point cloud registration further significantly impacts SLAM system precision; while a substantial volume of points awaiting registration enhances accuracy, it imposes considerable efficiency burdens upon the system. To address these challenges, this paper proposes a tightly coupled LiDAR/IMU SLAM system. This system achieves outstanding accuracy and efficiency through an enhanced Nano-GICP algorithm and graph optimisation. The core innovation lies in introducing a novel point cloud registration strategy: rather than searching for correspondences within downsampled point clouds, we employ NanoFLANN to construct a k-d tree on the original high-resolution point cloud. This is combined with Fsat-GICP to perform efficient nearest neighbour searches. Through false match rejection and iterative optimisation, we achieve higher accuracy with fewer registration points, thereby enhancing both computational speed and registration precision. Furthermore, we implement a tightly coupled factor graph optimisation backend that integrates IMU pre-integration, lidar odometry, and loop factor, significantly reducing drift. Extensive evaluation on the KITTI dataset and proprietary data demonstrates that our approach reduces absolute trajectory error by 55.61%, 47.32%, and 45.22% respectively compared to LEGO-LOAM, LIO-SAM, and FAST-LIO2, while maintaining real-time performance.</p>

错误：搜索内容不能为空，请输入英文关键词
错误：关键词超出字数限制，请精简
高级检索

Nano-GICP tightly-coupled LIO SLAM with graph optimization

  • 期刊论文

  • 出版模式：

    开放获取

  • 发表日期：

    2026年4月28日

查看全文
  • Qingyong Zhang,
  • Zhennan Lin,
  • Ruilin Dong,
  • Qunxiong Zheng,
  • Chengye Lin

摘要

Simultaneous Localisation and Mapping (SLAM) stands as a pivotal technology for large-scale positioning and map construction. During mapping operations in expansive environments, mobile robots incur inherent motion errors alongside point cloud overlaps and drift, consequently diminishing mapping accuracy. Laser point cloud registration further significantly impacts SLAM system precision; while a substantial volume of points awaiting registration enhances accuracy, it imposes considerable efficiency burdens upon the system. To address these challenges, this paper proposes a tightly coupled LiDAR/IMU SLAM system. This system achieves outstanding accuracy and efficiency through an enhanced Nano-GICP algorithm and graph optimisation. The core innovation lies in introducing a novel point cloud registration strategy: rather than searching for correspondences within downsampled point clouds, we employ NanoFLANN to construct a k-d tree on the original high-resolution point cloud. This is combined with Fsat-GICP to perform efficient nearest neighbour searches. Through false match rejection and iterative optimisation, we achieve higher accuracy with fewer registration points, thereby enhancing both computational speed and registration precision. Furthermore, we implement a tightly coupled factor graph optimisation backend that integrates IMU pre-integration, lidar odometry, and loop factor, significantly reducing drift. Extensive evaluation on the KITTI dataset and proprietary data demonstrates that our approach reduces absolute trajectory error by 55.61%, 47.32%, and 45.22% respectively compared to LEGO-LOAM, LIO-SAM, and FAST-LIO2, while maintaining real-time performance.