Time Efficient Solution for Formula Student Driverless Competition: A Unmanned Aerial Vehicle Scouting Approach

Abstract

Formula Student Driverless (FSD) is a famous self-driving race car competition in which the participating autonomous cars race on an unknown track. Many race cars, including the 2018 and 2019 champions, operate with a two-stage approach. The first stage is the training stage, which is used by the cars to observe the track information; the second stage is the execution stage, in which the cars move at full speed based on the information obtained in the first stage. However, a major limitation of this approach is that the cars have to move slowly during the training stage, since they need to gradually learn the track information and reserve enough time (e.g., 2 seconds) ahead of the operation to avoid collision. In addition to the above issue, previous cars are based on algorithms that are cannot be timely executed, which causes large operational delay and increases the risk of collision. To overcome these limitations, this paper presents a novel framework to enhance the performance of race cars. Specifically, the car is guided by a scouting unmanned aerial vehicle (UAV) that obtains the global track information with a monocular camera at the training stage. To implement the proposed framework, a set of algorithms are proposed to support various functionalities, including perception, simultaneous localization and mapping (SLAM), and path planning. Moreover, the proposed algorithms are highly time efficient, which can adapt to the environment at a faster rate than existing methods, thus supporting timely operation of cars and reducing the risk of collision. Our test results indicate that, with the proposed approach, the race car can obtain global trace information 50 seconds before the car reaches the finish line, which enables the race car to safely achieve a better racing performance.