LiDAR SLAM

Using a lidar's scanned point clouds to localize a robot and build a map of the environment at the same time.

Lidar SLAM is simultaneous localization and mapping that uses lidar as its primary sensor: as the robot moves, it aligns each new frame's point cloud with the existing map to work out its own position, while stitching the new point cloud into the map. Lidar measures distance directly and isn't affected by darkness, making it more stable than visual SLAM in large scenes and low-light places, though it lacks color and texture. 2D lidar SLAM is common in vacuums and warehouse AGVs; Google's open-source Cartographer supports both 2D and 3D. In 3D, Carnegie Mellon's 2014 LOAM splits the problem into a high-frequency odometry step and a low-frequency mapping step, and the MARS Lab at the University of Hong Kong's FAST-LIO2 further fuses in an IMU, running at up to 100 Hz and working well with narrow-field-of-view solid-state lidars too.

ExampleMounting a 3D lidar on a quadruped's back and driving it remotely around a campus loop, FAST-LIO2 builds a point-cloud map; afterward, during autonomous navigation, matching the live point cloud against this map tells the robot where it is.

Also called

Lidar-Based SLAM

Related

Simultaneous Localization and Mapping · LiDAR · LiDAR-Inertial Odometry · FAST-LIO2 · Visual SLAM · Loop Closure Detection

Sources

LOAM: Lidar Odometry and Mapping in Real-time (RSS 2014)
FAST-LIO2: Fast Direct LiDAR-inertial Odometry (arXiv:2107.06829)
Cartographer 官方文档 (Chinese)