Cooperative, connected and automated vehicles into a roundabout

Cooperative, connected and automated vehicles (CCAVs) are being constantly developed. Their introduction on public roads will be gradual, with a transition period where both automated and human-driven vehicles will share the same infrastructure. In order to investigate the interaction of CCAVs and traditional vehicles, this paper proposes a co-simulation between a microscopic traffic simulator and a dynamic driving simulator, introducing a real human-in-the-loop. The reference scenario is a single-lane roundabout. The behaviour of CCAVs is controlled by means of an artificial intelligence policy, based on reinforcement learning and trained to minimize the time needed for the vehicles to go through the roundabout. Drivers involved in the preliminary tests are asked to compare two scenarios with different percentages of CCAVs, in terms of traffic smoothness and safety feeling. The participants seem to prefer the a situation where the majority of traffic actors are CCAVs.