Gates n' poses: on the ai-way to speed of perception in autonomous racing drones

This thesis explores the application of AI in the realm of autonomous racing drones. The racing industry has historically driven the evolution of robust and efficient systems, with the recent focus shifting towards self-driving mechanisms. The challenges in this domain are complex, requiring precise control and perception systems with minimal reaction times. The research conducted at the Autonomous Robotics Research Center of the Technology Innovation Institute is presented, with a particular emphasis on perception systems for autonomous racing drones. The study introduces an innovative high-speed dataset and a pioneering approach for gate pose estimation powered by state-of-the-art keypoints detection neural network. For the purpose of evaluation, the research also presents an algorithm capable of reconstructing the map of an unknown track, leveraging the results provided by the gate pose estimation. This work contributes significantly to the field, enhancing the accuracy of perception systems and significantly lowering their computational complexity