// the find
gaoxiang12/faster-lio
Faster-LIO: Lightweight Tightly Coupled Lidar-inertial Odometry using Parallel Sparse Incremental Voxels
Faster-LIO is a ROS1 lidar-inertial odometry package built on top of HKU-Mars' FastLIO2, swapping FastLIO2's ikd-tree for a custom incremental voxel structure (iVox) to get a 1.5-2x throughput increase. It's aimed at robotics researchers doing real-time SLAM on solid-state or spinning lidars who need headroom on commodity CPUs rather than a GPU.
The performance claim isn't marketing fluff - the README ships actual FPS comparisons against FastLIO2 on AMD/Intel CPUs using the NCLT dataset, and the core contribution (iVox, with linear and PHC variants selectable at compile time) is a real, parallelizable replacement for the ikd-tree nearest-neighbor lookup, not a cosmetic wrapper. Hardware support is broad out of the box - Livox (avia, horizon, mid360), Ouster, Hesai, RoboSense and Velodyne configs are all included. There's a peer-reviewed RA-L 2022 paper backing the method, included as a PDF in the repo.
ROS1 only (melodic/noetic) with no ROS2 port, which is a real problem now that ROS1 is EOL. The maintainers' own README admits the original dataset links are dead and routes you to Baidu Yun with an access code - a rough path for anyone outside China. The known-issues section calls out a core dump from -march=native on specific gcc/CPU combos and says iVox is 'sensitive' to voxel size without giving real tuning guidance, meaning you'll be debugging stability by trial and error per sensor. It also vendors a chunk of FastLIO2's IKFoM toolkit plus a bundled TBB tarball for old Ubuntu, so you inherit someone else's dependency pinning instead of using your package manager.