A Tightly Coupled Localization System for High-Speed Autonomous Racing in GNSS-Denied Environments

Accurate and reliable localization remains a critical challenge in autonomous racing. Unlike urban driving, high-speed racing involves frequent turns and rapid acceleration under extreme conditions, where severe vibrations and thermal stress can degrade sensors or interrupt signals. This article presents a multisensor fusion localization system that maintains positioning accuracy up to 250 km/h, even during complete global navigation satellite system (GNSS) outages. The system uses a velocity-forward error-state Kalman filter (ESKF) that integrates optical ground-speed measurements and inertial angular rates from an inertial measurement unit (IMU). In the mapping phase, GNSS constraints are used to build the main track map. For GNSS-denied underground areas, a loop-closure method based on natural geometric constraints at tunnel entrances and exits corrects cumulative drift. During online localization, efficient local map retrieval and a dynamic keyframe update mechanism ensure real-time performance. Extensive experiments on a Super Formula race car show that the proposed system achieves high precision, stable localization on both surface and underground tracks, outperforming existing methods in accuracy and robustness. The system has been successfully deployed on the Abu Dhabi Autonomous Racing League (A2RL) autonomous racing platform.