Scholarlink

<p>Simultaneous localization and mapping (SLAM) is widely regarded as an effective approach for enabling autonomous navigation and localization guidance in harsh or unstructured environments. However, existing vision SLAM or LiDAR SLAM methods often suffer from significant trajectory drift when traversing uneven terrain, with errors particularly pronounced along the vertical axis. To address these challenges, this paper presents a SLAM method that integrates LiDAR and inertial measurement unit (IMU) data, aiming to enhance pose estimation accuracy in unstructured environments while reducing computational load. The proposed approach first employs a hash map-based data structure to remove outliers from the input point cloud, ensuring that the number of points within most voxels meets algorithmic requirements. Building upon this, an adaptive voxel mechanism based on point cloud density is introduced. The global voxel size is dynamically adjusted by using the average point cloud density, enabling more effective preservation of geometric features in the environment. Subsequently, LiDAR keyframes are adaptively generated by combining environmental width and pose estimation from inertial odometry. Once a keyframe is selected, the system performs pose estimation and registration. Experimental results demonstrate that the proposed method achieves higher pose estimation accuracy than current advanced LiDAR SLAM frameworks in unstructured environments and exhibits lower CPU consumption compared to leading LIO algorithms.</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

AVK-SLAM: an efficient LiDAR–inertial SLAM with density-aware voxelization and keyframe selection in unstructured environments

  • 期刊论文

  • 发表日期：

    2026年2月27日

查看全文
  • Shuai Liu,
  • Xiuzhi Li

摘要

Simultaneous localization and mapping (SLAM) is widely regarded as an effective approach for enabling autonomous navigation and localization guidance in harsh or unstructured environments. However, existing vision SLAM or LiDAR SLAM methods often suffer from significant trajectory drift when traversing uneven terrain, with errors particularly pronounced along the vertical axis. To address these challenges, this paper presents a SLAM method that integrates LiDAR and inertial measurement unit (IMU) data, aiming to enhance pose estimation accuracy in unstructured environments while reducing computational load. The proposed approach first employs a hash map-based data structure to remove outliers from the input point cloud, ensuring that the number of points within most voxels meets algorithmic requirements. Building upon this, an adaptive voxel mechanism based on point cloud density is introduced. The global voxel size is dynamically adjusted by using the average point cloud density, enabling more effective preservation of geometric features in the environment. Subsequently, LiDAR keyframes are adaptively generated by combining environmental width and pose estimation from inertial odometry. Once a keyframe is selected, the system performs pose estimation and registration. Experimental results demonstrate that the proposed method achieves higher pose estimation accuracy than current advanced LiDAR SLAM frameworks in unstructured environments and exhibits lower CPU consumption compared to leading LIO algorithms.