FAST-LIO2: Fast Direct LiDAR-Inertial Odometry

This article presents FAST-LIO2: a fast, robust, and versatile LiDAR-inertial odometry framework. Building on highly efficient tightly coupled iterated Kalman filter, FAST-LIO2 has two key novelties that allow accurate LiDAR navigation (and mapping). The first one is directly registering raw points to the map subsequently update map, i.e., mapping) without extracting features. enables exploitation of subtle features in environment and, hence, increases accuracy. elimination hand-engineered feature extraction module also makes it naturally adaptable emerging LiDARs different scanning patterns; second main novelty maintaining by an incremental k-dimensional (k-d) tree data structure, k-d ( ikd-Tree ), updates (i.e., point insertion delete) dynamic rebalancing. Compared with existing structures (octree, R $^\ast$ -tree, xmlns:xlink="http://www.w3.org/1999/xlink">nanoflann tree), achieves superior overall performance while supports downsampling tree. We conduct exhaustive benchmark comparison 19 sequences from variety open datasets. consistently higher accuracy at much lower computation load than other state-of-the-art systems. Various real-world experiments solid-state small field view are conducted. Overall, computationally (e.g., up 100 Hz mapping large outdoor environments), robust reliable pose estimation cluttered indoor environments rotation 1000 deg/s), applicable both multiline spinning LiDARs, unmanned aerial vehicle (UAV) handheld platforms, Intel- ARM-based processors), still achieving methods. Our implementation system structure open-sourced Github.