Cooperative, connected and automated mobility (CCAM) is one of the topics of the European Project AI@EDGE. The aim is to develop a framework in which artificial intelligence (AI), 5G and edge computing are combined to allow smooth flow of traffic into roundabouts. Cooperative, connected and automated vehicles (CCAVs) are being constantly developed with the aim of their widespread utilization on public roads in the near future. The introduction of CCAVs will be gradual with a transition period when both automated and human driven vehicles will be present on the road. In this situation, the understanding of CCAVs interaction with traditional human-driven cars is of paramount importance to assess their safety and functionality. On road tests are prohibitive for the lacking of proper CCAVs, costs and safety issues. In this situation, commercial traffic simulators software may represent a viable alternative, since they can reproduce most traffic conditions, allowing to understand traffic dynamics in the pertinent situations. However, these tools rely on driver models that can represent the human behaviour only in a very approximated way, thus limiting their ability to reliably simulate CCAVs and human drivers interactions. To overcome these limitations, this paper presents a co-simulation between a traffic simulator software and a dynamic driving simulator. This coupling allows to introduce a human-in-the-loop and to evaluate the effect of a real human driver in the simulated environment. The proposed approach has been developed as a use case of a comprehensive research project, named AI@EDGE, funded by European Commission, focusing on the improvement of 5G networks through artificial intelligence and edge computing. In particular, this use case has the objective of understanding the requirements, benefits and limitations of edge computing and artificial intelligence in traffic management. With this aim in view, a critical aspect is represented by interactions between human drivers and CCAVs. In this context, the use of a dynamic driving simulator can represent a powerful tool to investigate the interactions between human drivers and automated cars. In fact, it allow to test the AI algorithm in a safe, controllable and repeatable environment, avoiding costs and safety issues related to testing in the real traffic. Additionally, by this approach, testing time can be greatly reduced as different situations can be easily reproduced by modifying the parameters of the simulations. As reference traffic scenario, a single-lane roundabout has been selected. The roundabout is negotiated by simulated vehicles, both automated and human-driven, and an actual human driven vehicle. The latter is controlled by a human driving the dynamic driving simulator. The driving simulator employed for the research is a cable driven dynamic simulator located at DriSMi laboratory of Politecnico di Milano. The dynamics of the vehicle driven in the driving simulator is simulated by a complete real-time multi-body vehicle model. The driver on the driving simulator, beside the motion feedback of the simulator, is also immersed in a graphical and sound environment providing a fully immersive and realistic experience. The traffic in the roundabout is simulated using an open source software for microscopic and continuous traffic simulations, widely employed for reproducing a wide range of traffic scenarios. In order for the automated vehicles to negotiate the roundabout, a reinforcement learning (RL) algorithm is developed. Traffic simulation and RL algorithm run in parallel to the driving simulator environment, while a real time scheduler synchronizes the different simulations. Preliminary tests are carried on considering a panel of different drivers. In order for them to have a complete perception of the traffic scenario, drivers repeat the test entering the roundabout one time for each of its legs. For each leg, the manoeuvre is repeated with different percentages of automated vehicles. After the test, drivers are asked to fill a survey to evaluate their perceptions in the different situations. The responses of the testers are evaluated to understand their perception of the CCAVs behaviour. In particular, the acceptance of the human drivers with respect to the presence of CCAVs is investigated with reference to drivers’ perception of traffic fluidity and safety. Drivers’ preferences of higher or lower CCAVs percentage in the considered traffic conditions are also analysed. From the results, it seems that human drivers accept driving together with CCAVs.