Coordination of Automated Vehicles at Intersections: Decision, Efficiency and Control

This papers studies the kind of control that is needed to efficiently coordinate multiple automated vehicles. An intersection is chosen in order to present the main concept but consequences of this work also hold for other areas of cooperation, such as lane changes or maneuvers in parking lots. We chose the classical framework for multi-robots systems: the coordination space i.e. we assume the future paths are known and fixed. The problem is to coordinate the speeds of the vehicles. We first prove a theorem stating that a smooth feedback control cannot always avoid gridlocks: for more than 2 vehicles, there are always starting states ending into gridlocks. The paper then proposes some ways to avoid this drawback, leading to a better conceptual way to take decision in such a cooperative system, in order to have provable efficient decision and control.