// the take
FAST-LIVO2 is a tightly-coupled LiDAR-inertial-visual odometry and mapping system out of HKU-MARS, built for real-time robot localization and 3D reconstruction in environments where LiDAR or vision alone degrades (dark, textureless, or geometrically sparse spaces). It's for robotics researchers building SLAM stacks on ROS1, not application developers looking for a drop-in library.
Peer-reviewed (T-RO 2024) with a dedicated resource-constrained-platforms paper, so there's real validation behind the claims beyond a README. Uses a direct (non-feature-extraction) visual-inertial formulation fused with LiDAR, which is noticeably cheaper computationally than feature-based VIO pipelines. Ships a companion calibration tool (FAST-Calib) and open hardware design (CAD, STM32 sync firmware, ROS driver) so you can actually build a compatible sensor rig instead of guessing at extrinsics.
Locked to ROS1 on Ubuntu 18.04-20.04 via catkin, with a pinned Sophus commit (a621ff) and a separate Vikit dependency — classic academic dependency hell, no ROS2 port, no Docker image. Bundles Livox's proprietary ros_driver headers directly in include/, so out of the box it's coupled to that one LiDAR vendor. Licensed GPLv2 with commercial use requiring a separate negotiated license from the authors, which kills casual commercial adoption. No CI, no tests, and support is literally 'email the PhD student' — fine for a research baseline, risky as a dependency in anything that needs to keep working.
We read the README, the directory tree and a key source file in October 2026. Nobody pays to be listed. How we judge →