FLAGDroneRacing: An Autonomous Drone Racing System.

The article presents FLAGDroneRacing, a comprehensive autonomous drone racing system developed by researchers at the Beijing Institute of Technology. The system integrates a loosely coupled visual-inertial localization method combining binocular vision and inertial measurement units (IMUs) for robust real-time pose estimation in GPS-denied environments, a lightweight CPU-based perception module for autonomous mapping and gate detection using Hough circle detection, and a real-time trajectory planner employing an enhanced A* pathfinding algorithm and B-spline-based optimization to generate smooth, collision-free, and dynamically feasible trajectories. The system operates entirely onboard a lightweight drone equipped with an Intel NUC computer and stereo cameras, achieving high-frequency localization (up to 40 Hz), mapping (20 Hz), and planning (20 Hz) without reliance on external infrastructure. Tested in both simulation and indoor racing environments, FLAGDroneRacing demonstrates effective autonomous navigation through complex courses, though current limitations include suboptimal trajectory tracking at speeds above 5 m/s and limited handling of fast-moving dynamic obstacles, which are identified as areas for future improvement.