Model predictive control for autonomous and cooperative driving. (Commande prédictive pour conduite autonome et coopérative)

Autonomous driving has been gaining more and more attention in the last decades, thanks to its positive social-economic impacts including the enhancement of traffic efficiency and the reduction of road accidents. A number of research institutes and companies have tested autonomous vehicles in traffic, accumulating tens of millions of kilometers traveled in autonomous driving. With the vision of massive deployment of autonomous vehicles, researchers have also started to envision cooperative strategies among autonomous vehicles. This thesis deals with the control architecture design of individual autonomous vehicles and cooperative autonomous vehicles. Model Predictive Control (MPC), thanks to its efficiency and versatility, is chosen as the building block for various control architectures proposed in this thesis. In more detail, this thesis first presents a classical hierarchical control architecture for individual vehicle control that decomposes the controller into a motion planner and a tracking controller, both using nonlinear MPC. In a second step, we analyze the inability of the proposed planner in handling logical constraints raised from traffic rules and multiple maneuver variants, and propose a hybrid MPC based motion planner that solves this issue. We then consider the convoy control problem of autonomous vehicles in which multiple vehicles maintain a formation during autonomous driving. A hierarchical formation control architecture is proposed composing of a convoy supervisor and local MPC based vehicle controllers. Finally, we consider the problem of coordinating a group of autonomous vehicles at an intersection without traffic lights. A hierarchical architecture composed of an intersection controller and multiple local vehicle controllers is proposed to allow vehicles to cross the intersection smoothly and safely.