In context
In mid-2024, autonomous navigation robots were increasingly deployed in unstructured environments such as forests, mountains, and construction sites. However, conventional LiDAR-inertial SLAM algorithms often suffered from accuracy degradation, localization drift, or failure under intense motion and rough terrain, limiting their industrial applicability.
What was reported
Researchers from Harbin Institute of Technology (Weihai) and partners proposed a multi-sensor fusion SLAM algorithm tailored for complex terrain. The method addresses severe point cloud distortion during aggressive motion by introducing an adaptive sub-frame segmentation technique. Radar frames are dynamically divided based on IMU-measured motion intensity, and IMU pre-integration compensates for distortion within each sub-frame, improving robustness.
The front-end employs an iterative error-state Kalman filter (IESKF) to fuse LiDAR and IMU data, providing accurate initial poses. The back-end uses factor graph optimization integrating LiDAR-inertial odometry, loop closure, and GPS factors to enhance global consistency.
Tests in intense motion, campus, and forest scenarios showed that the algorithm outperformed FAST-LIO2 and LIO-SAM in localization accuracy, mapping clarity, and robustness under intense motion.
Why it mattered
This work offered a practical solution for reliable SLAM in demanding environments, potentially extending the operational envelope of autonomous ground robots in sectors like construction, agriculture, and field inspection, where terrain-induced motion disturbances are common.
“The proposed method achieves higher localization accuracy, clearer mapping, and greater robustness in intense motion scenes.”
Source: 《机器人》期刊 (robot.sia.cn) · Published 2024-07-20 · “复杂地形环境下的多传感器融合SLAM技术”