Scholarlink

<p>Traditional SLAM methods are often not sufficient to meet the accuracy requirements of localization and mapping in advanced mining applications. In response to this challenge, this paper proposes a tightly coupled 3D LiDAR-inertial SLAM method. The proposed algorithm adopts IMU forward propagation to compute the robot’s current position. It incorporates preprocessed LiDAR feature points, wheel odometry, and IMU data through the use of an iterative error-state Kalman filter. Motion distortion of the LiDAR data is corrected by backward propagation, which properly compensates for the cumulative error from the IMU data. The back-end uses the inter-frame pose transformation matrix from the LiDAR front-end to generate the relative pose constraint factors. These are combined with loop closure detection factors to create a tightly combined global optimization paradigm. Factor graph optimization was employed to enhance the accuracy of pose estimation. Open-source and field datasets collected from an open-pit tailing pond were employed to validate the robustness and accuracy of the algorithm. The outcome suggests that in an open-channel environment in the tailings pond, spanning over 160 meters, we propose algorithm attained 37.83% better accuracy compared to fast_lio2 and 51.04% better accuracy compared to sc-lego-loam, with 1.545, 2.453, and 3.156 errors, and mapped successfully in a complex environment.</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

A Tightly Coupled LiDAR-IMU Based 3D Mapping Method for Mining Robots

  • 期刊论文

  • 发表日期：

    2026年6月29日

查看全文
  • Mingqi Liang,
  • Mudassir Mehmood,
  • Tianqiang Zhu,
  • Yidu Hong,
  • Ning Pang,
  • Shuwen Zhang,
  • Wen Nie

摘要

Traditional SLAM methods are often not sufficient to meet the accuracy requirements of localization and mapping in advanced mining applications. In response to this challenge, this paper proposes a tightly coupled 3D LiDAR-inertial SLAM method. The proposed algorithm adopts IMU forward propagation to compute the robot’s current position. It incorporates preprocessed LiDAR feature points, wheel odometry, and IMU data through the use of an iterative error-state Kalman filter. Motion distortion of the LiDAR data is corrected by backward propagation, which properly compensates for the cumulative error from the IMU data. The back-end uses the inter-frame pose transformation matrix from the LiDAR front-end to generate the relative pose constraint factors. These are combined with loop closure detection factors to create a tightly combined global optimization paradigm. Factor graph optimization was employed to enhance the accuracy of pose estimation. Open-source and field datasets collected from an open-pit tailing pond were employed to validate the robustness and accuracy of the algorithm. The outcome suggests that in an open-channel environment in the tailings pond, spanning over 160 meters, we propose algorithm attained 37.83% better accuracy compared to fast_lio2 and 51.04% better accuracy compared to sc-lego-loam, with 1.545, 2.453, and 3.156 errors, and mapped successfully in a complex environment.