Design of an autonomous vehicle's lateral controller for high speed and emergency maneuvers

Autonomous cars are no longer science fiction. As I am writing many car manufacturers are already testing their self-driving systems on road, meaning that soon our cars will ask us: where do you want me to drive you? This is not just a marketing strategy but it is dictated by the necessity of introducing the ultimate automatic safety control on board of our main mean of transport. In 2016 - despite the regulations on safety systems as ESP and ABS - still too many people lose their life in car accidents, leading to the conclusion that humans’ distractions are to be blamed. Setting up a fully autonomous vehicle requires many competences which range from computer vision to systems and control. This thesis focuses on the design of a lateral controller capable of engaging corners and emergency maneuvers at high speed. The work is supposed to be a further development of Gimondi-Tiralongo, 2015 whose solution did not proved to be capable of facing some scenarios which may occur in everyday driving. The first part of the dissertation is devoted to framework description and theoretical modeling of the vehicle dynamics. Furthermore, the error signals with respect to a target trajectory are defined by extending the state-space system. Eventually, an overview of the Linear Parameter Varying (LPV) theory is introduced since the system under analysis is strongly dependent on longitudinal velocity. The control strategy presented in Gimondi-Tiralongo, 2015 is then analyzed in detail to highlight strengths and weaknesses. The main downfall is the concept of Look-Ahead distance which is used to get a preview on the upcoming corner, and shows a lack of robustness in terms of adaptation to different maneuvers. From these conclusions, a new controller has been derived by exploiting the knowledge on future road curvature: the solution consists in embedding a yaw rate tracker inside an outer loop which feedbacks the lateral error. The improvements brought by the new control strategy are eventually validated on CarSim in a series of index maneuvers which take the vehicle at the limits of handling. The project is sponsored by Magneti Marelli which made available a fully autonomous vehicle and a private circuit used to validate the controllers in a real scenario. Data gathered during the experimental sessions are reported and compared to the results obtained via simulation.