hku-mars/ikd-tree

Robots that use LiDAR sensors need to constantly manage massive streams of 3D points representing their surroundings. ikd-Tree is a tool that helps robots organize and query this spatial data efficiently. Instead of rebuilding a complete map of points every time new sensor data arrives, it incrementally updates the existing structure, saving significant computation time. At a high level, it works like a living filing cabinet for 3D points. You can add new points as the robot moves, delete points within specific rectangular regions, and search for the closest points to any given location. One notable design choice is how it handles deletions: rather than immediately removing points, it marks them as deleted in the background and cleans them up later during a rebuilding phase. This asynchronous approach keeps the robot responding quickly without getting stuck on cleanup tasks. The primary users are roboticists and researchers building autonomous systems, particularly those working with LiDAR-inertial odometry, which is how a robot figures out where it is by combining laser scans with motion sensors. The project was developed alongside FAST-LIO2, a system for fast robot navigation. Anyone building autonomous vehicles, drones, or mapping systems that need to rapidly query nearby obstacles would find this useful. The project is written in C++ for performance and is released under the GPLv2 license, meaning it is free for research and open-source use, but commercial applications require contacting the developers. The repository includes demo programs that visualize the search and delete functions, which helps illustrate how the tool manages 3D point clouds in practice.