FAST-LIVO2 is a LiDAR-inertial-visual odometry system that fuses data from LiDAR, inertial measurement units, and cameras to perform real-time localization and 3D reconstruction.
The system addresses the challenge of accurate robot localization and mapping in degraded environments where individual sensors may fail or perform poorly. It combines direct LiDAR-inertial-visual fusion to achieve both efficiency and accuracy. The approach tightly couples measurements from all three sensor modalities to produce robust odometry estimates and reconstruct 3D scenes, including colored point clouds and mesh representations.
The tool suits roboticists and researchers working on autonomous systems that operate in challenging conditions such as low-light, GPS-denied, or visually featureless environments. It is designed for onboard robotic deployment and real-time performance. The system includes support for Gaussian splatting and NeRF-based reconstruction methods, enabling advanced 3D scene representation beyond traditional point cloud mapping. Users should be aware that the tool requires careful sensor synchronization and calibration; the developers provide open-source hardware designs for a synchronized handheld device and recommend their FAST-Calib toolkit for LiDAR-camera extrinsic calibration.
The project maintains active engagement with the research community through publication in a top-tier robotics journal and provides comprehensive supporting materials including evaluation datasets, hardware designs with CAD files and firmware, and accompanying video demonstrations. Development includes investigation of deployment on resource-constrained platforms, extending the system's applicability beyond high-end computing environments.
