Design of a perception system for the Formula Student Driverless competition: from vehicle sensorization to SLAM

Formula Student Driverless is an international racing competition held among universities, where the vehicles must complete a set of trials without any human intervention. Together with RaceUP, the Formula Student team of the University of Padova, this thesis represents the beginning of the project to build an autonomous prototype to compete in the Driverless Cup in the 2024 season. Three important aspects of an autonomous system design will be tackled: vehicle sensorization, perception, and simultaneous localization and mapping (SLAM), with the main focus on the development of the last one. The proposed approach for the back-end is based on the optimization of a factor graph, holding information about car poses and landmarks positions, by exploiting spatial and kinematic constraints between its vertices. The full back-end pipeline has been tested thoroughly, step by step, allowing to obtain satisfactory results on the different virtual tracks used for testing. Using both modern and classical techniques, we can process information produced by the stereo camera and the LIDAR, to be able to localize the colored cones delimiting the track. The estimation of cones positions serves then as input for other important modules of the car, such as the control part and the SLAM pipeline. Finally, a complete dataset has been acquired by properly sensorizing RaceUP's last year's car: having real data represents a helpful resource to make experiments and validate the system, even without the availability of the actual vehicle prototype.