Formula Student competitions are worldwide events where students design, build and race innovative formula style vehicles. In order to remain at the cutting edge of technology, some of these competitions such as the Formula Student Germany (FSG) competition now requires driverless capabilities from the vehicles. This project serves as a proof of concept for the software systems required to implement an autonomous vehicle for these competitions. A modular system is designed to allow
for individual development of components, and for future improvements on these individual systems. These systems include perception, mapping and path planning, state estimation and controls.

The perception uses a stereoscopic camera in order to identify cones using the You Only Look Once (YOLO) algorithm, and estimates their position with a disparity map. Mapping and path planning uses a Fast Simultaneous Localization And Mapping (SLAM) algorithm to map the cones and a Rapidly Exploring Random Tree (RRT) to create waypoints for a path. State estimation uses an Extended Kalman Filter (EKF) to fuse vehicle sensor data. Finally, the control systems uses a pure-pursuit controller to drive the vehicle. Each module is implemented as a Robot Operating System (ROS) node to allow for modular design and improvements on individual system components.

The perception is individually tested using a physical stereo camera, and in a simulation environment. All other aspects are tested in simulation. Finally, the full system is integrated in an end-to-end simulation and tested in acceleration, skidpad and autocross events which it successfully completed. The system also undergoes Hardware-In-the-Loop (HIL) testing with the autonomous system running on an Nvidia Jetson TX2 and also completes these same events. A successful driverless system is developed and tested through simulations and HIL tests meeting 13 out of 14 performance metrics.