Tightly coupled LiDAR–inertial–visual odometry built on FAST-LIVO2, extended with a pose-graph optimization (PGO) hook: the front end keeps the FAST-LIVO2 EKF estimate while loop-closure results are fed back over ROS for state correction and optional local map rebuild. SC is intended to be used together with Scan Context-style place recognition (actual loop detection and factor-graph solving live in an external PGO package).
Publication: Add your title, venue, and arXiv/DOI here when available.
This project extends FAST-LIVO2: Fast, Direct LiDAR-Inertial-Visual Odometry from HKU Mars Lab. Use the official repository for the original implementation, datasets, and build prerequisites (ROS, PCL, Eigen, OpenCV, Vikit, Sophus, etc.):
https://github.com/hku-mars/FAST-LIVO2
- FAST-LIVO2 front end: direct VO + voxel-map LiDAR + IMU; Livox and spinning LiDAR configs supported.
- PGO integration: subscribes to optimized keyframe poses/IDs; periodically updates the filter state and can rebuild the local map.
- Dataset launches: AVIA, NTU VIRAL, HILTI 2022, MARS LVIG, and more under
launch/. - Scripts: meshing and COLMAP export helpers in
scripts/(optional, environment-dependent).
| Item | Suggested |
|---|---|
| ROS | ROS Noetic (Ubuntu 20.04) or another ROS1 distro you validate |
| C++ | C++17 |
| Libraries | Eigen3, PCL, OpenCV, Boost (thread), OpenMP optional |
Catkin package name: fast_livo (see CMakeLists.txt / package.xml).
ROS / catkin (install or clone into the same workspace):
roscpp,sensor_msgs,geometry_msgs,nav_msgs,tf,tf2,tf2_rosimage_transport,cv_bridge,pcl_rosvikit_common,vikit_roslivox_ros_driverwhen using Livox hardware/drivers
Full loop closure (optional):
mapping_with_pgo.launchincludesaloam_velodyne'sfastlivo2_pgo.launch. Build that package in your workspace or run your own PGO/back-end nodes with matching topic names.
# Clone into ~/catkin_ws/src as fast_livo so roslaunch finds the package
cd ~/catkin_ws/src
git clone <YOUR_REPO_URL> fast_livo
cd ~/catkin_ws
rosdep install --from-paths src --ignore-src -r -y # if applicable
catkin_make -DCMAKE_BUILD_TYPE=Release
source devel/setup.bashroslaunch fast_livo mapping_avia.launchroslaunch fast_livo mapping_with_pgo.launch| Launch file | Purpose |
|---|---|
mapping_avia.launch |
Livox AVIA + config/avia.yaml |
mapping_avia_direct.launch |
AVIA variant |
mapping_avia_marslvig.launch |
MARS LVIG |
mapping_hesaixt32_hilti22.launch |
HILTI 2022 |
mapping_ouster_ntu.launch |
NTU VIRAL (Ouster) |
mapping_with_pgo.launch |
Front end + fastlivo2_pgo.launch + RViz |
Edit topic names and extrinsics in config/*.yaml for your bag or robot.
| Direction | Topic / note |
|---|---|
| Publish | /cloud_registered_local (sensor_msgs/PointCloud2, frame_id: camera_init) |
| Subscribe | /aft_pgo_path (nav_msgs/Path) — optimized keyframe poses |
| Subscribe | /key_frames_ids (std_msgs/Header, keyframe id in seq) |
Environment variable SC_FASTLIVO2 is set in mapping_avia.launch to this package's path for external nodes.
Additional parameters are read in LIVMapper (defaults in code if omitted):
pgo_integration_enable,pgo_update_frequency,pgo_state_update_enable,pgo_map_rebuild_enablepgo_update_weight,pgo_cov_weightmax_keyframe_num,keyframe_meter_gap,keyframe_deg_gap,surrounding_search_radius
See readParameters and PGO callbacks in include/LIVMapper.h and src/LIVMapper.cpp.
fast_livo/
├── CMakeLists.txt
├── package.xml
├── assets/
├── config/
├── include/
├── launch/
├── rviz_cfg/
├── scripts/
└── src/
The odometry and mapping front end follows FAST-LIVO2 (hku-mars/FAST-LIVO2). Please cite the FAST-LIVO2 paper and credit the upstream authors in any derivative work; clearly describe additions such as PGO integration and your loop-closure back end.
This project is licensed under the GNU GPLv2; see LICENSE. Maintainer metadata in package.xml may follow upstream templates; the LICENSE file prevails.