FAST-LIO for LASER

fast_lio estimates vehicle motion and builds a map from LiDAR scans and IMU measurements. This LASER fork provides the ROS 2 estimator used by the LUS LiDAR-odometry integration, including scan-corrected and high-frequency odometry outputs.

FAST-LIO2 uses an iterated error-state Kalman filter and incremental mapping with ikd-Tree. It can register raw points without a separate feature-extraction stage.

Build and run

LUS supplies this repository through laser_fast_lio_core. With its dependencies available in a ROS 2 workspace:

colcon build --packages-up-to fast_lio
source install/setup.bash

Use the LASER wrapper to select a calibrated sensor profile:

export UAV_NAME=uav1
ros2 launch laser_fast_lio_core fast_lio.launch.py topic_pcl:=/uav1/livox/lidar topic_imu:=/uav1/livox/imu

To start the node through this package's own launch, pass a full configuration path:

ros2 launch fast_lio mapping.launch.py namespace:=uav1 config_file_path:=/absolute/path/to/sensor.yaml topic_pcl:=/uav1/livox/lidar topic_imu:=/uav1/livox/imu

The launch uses config_file_path; the older upstream config_file and config_path examples do not match this fork's launch arguments. use_sim_time defaults to true when REAL_UAV=false.

Interfaces

The supplied mapping launch remaps node topics as follows for uav1:

Direction	Topic	Type	Purpose
Input	topic_pcl	sensor_msgs/msg/PointCloud2	LiDAR scan.
Input	topic_imu	sensor_msgs/msg/Imu	IMU samples.
Output	/uav1/fast_lio/odometry	nav_msgs/msg/Odometry	Scan-corrected state.
Output	/uav1/fast_lio/odometry_high_freq	nav_msgs/msg/Odometry	High-frequency propagated state.
Output	/uav1/fast_lio/cloud_registered	sensor_msgs/msg/PointCloud2	Registered point cloud.
Output	/uav1/fast_lio/cloud_registered_body	sensor_msgs/msg/PointCloud2	Cloud in the body frame.
Output	/uav1/fast_lio/cloud_effected, /uav1/fast_lio/laser_map	sensor_msgs/msg/PointCloud2	Selected points and map output.
Output	/uav1/fast_lio/imu/out	sensor_msgs/msg/Imu	Estimator IMU output.
Output	/uav1/fast_lio/path	nav_msgs/msg/Path	Estimated path.
Service	/uav1/map_save	std_srvs/srv/Trigger	Saves the recorded map if PCD saving is enabled.

This implementation subscribes to PointCloud2. The Livox CustomMsg subscription in the source is commented out; configure the driver or simulator to publish the point fields expected by the selected preprocessing mode.

Configuration

The sensor YAML is passed directly to the node. LUS profiles live in laser_fast_lio_core/params/.

Parameter	Meaning
preprocess.lidar_type	2 selects the Velodyne handler, 3 Ouster and 4 Mid-360. Other values use the generic handler in this fork.
preprocess.timestamp_unit	Per-point time unit: 0 seconds, 1 milliseconds, 2 microseconds or 3 nanoseconds.
preprocess.scan_line, preprocess.scan_rate, preprocess.blind	Scan structure, nominal rate and minimum range.
feature_extract_enable, point_filter_num	Feature extraction and point thinning.
filter_size_surf, filter_size_map, cube_side_length, max_iteration	Surface/map resolution, local map size and filter iteration limit.
common.time_sync_en, common.time_offset_lidar_to_imu	Software synchronization and calibrated sensor time offset.
mapping.acc_cov, gyr_cov, b_acc_cov, b_gyr_cov	IMU measurement and bias noise settings.
mapping.extrinsic_T, mapping.extrinsic_R	LiDAR translation and rotation in the IMU frame.
mapping.extrinsic_est_en	Enables online extrinsic estimation.
mapping.fov_degree, mapping.det_range	Mapping field of view and range.
publish.*	Enables path, cloud and map outputs.
pcd_save.pcd_save_en, pcd_save.interval, map_file_path	Map recording, scan grouping and save path.
imu_in_gravity_unit	Acceleration is in g when true, m/s² when false.
velocity_in_body_frame	Selects the velocity output frame.
transform.enable, transform.enable_imu	Enables the configured TF handling.
transform.fcu_frame, world_frame, lidar_frame, imu_frame	Frame names within the transform group.

LiDAR and IMU timestamps must agree, and scans need usable per-point timing for motion compensation. Prefer hardware synchronization; use the software sync option only when external synchronization is unavailable. Supply measured extrinsics and the TF relationships for your sensor mounting.

Enable PCD saving before calling map_save. An interval of -1 retains all scans in one recording and can exhaust memory on long runs.

References and credits

• FAST-LIO2: Fast Direct LiDAR-inertial Odometry.
• FAST-LIO: A Fast, Robust LiDAR-inertial Odometry Package by Tightly-Coupled Iterated Kalman Filter.
• FAST-LIO2 video and FAST-LIO1 video.

Upstream contributors include Wei Xu, Yixi Cai, Dongjiao He, Fangcheng Zhu, Jiarong Lin, Zheng Liu and Borong Yuan. The ROS 2 fork credits Ericsiii. FAST-LIO builds on work including LOAM, Livox Mapping, LINS, Loam Livox and ikd-Tree.

The repository retains its GPL version 2 license and bundled dependency notices.