Design and experimental evaluation of a model-free controller for autonomous intersection crossing under imperfect localization

Despite the extensive amount of research that exists regarding Connected and Autonomous Vehicles (CAVs), experimental results are few. Furthermore, most works assume an accurate vehicle model and either assume perfect localization or ensure that the localization accuracy is very high by conducting overly specific experiments or utilizing sensor setups that are difficult to be replicated in real scenarios. Motivated by these facts, we study an unsignalized intersection crossing problem with fully autonomous traffic and develop an online, safety-critical and model-free centralized controller for CAVs under imperfect localization. The centralized controller follows a First-Come-First-Served (FCFS) reservation policy and ensures no more than one vehicle is present inside the center of the intersection in order to avoid collisions. Realistic experiments are conducted with 1:10 scaled vehicles and the results are discussed.