Autonomní řízení monopostu Formula Student v simulátoru

Abstract

This master's thesis describes the design and implementation of an autonomous driving system for a Formula Student Driverless race car within the CARLA simulator. The thesis first describes the competition rules, compares available simulators for autonomous driving and analyses the properties of sensors relevant for the detection of cones delimiting the race track. The implementation part covers the complete autonomous pipeline from cone detection in LiDAR point clouds, through colour classification by a convolutional neural network, track mapping using Graph SLAM, to vehicle control by the Pure Pursuit and Stanley controllers. The system operates in two modes corresponding to the DV Cup disciplines -- a reactive mode for the first lap of an unknown track and a racing mode that exploits the previously built map. The functionality of both modes is verified on three test tracks, and the results are compared in terms of lap time and trajectory tracking accuracy. The work also includes an evaluation of the achieved results and a comparison with existing approaches in Formula Student Driverless.