Future Roundabouts Relying on 5G, Edge Computing and Artificial Intelligence

Giorgio Previati, Elena Campi, Lorenzo Uccello, Antonino Albanese, Alessandro Roccasalva, Gabriele Santin, Massimiliano Luca, Bruno Lepri, Laura Ferrarotti, Nicola di Pietro, Marco Ponti, Gianpiero MastinuLecture Notes in Mobility 2025 · DOI · OpenAlex

Abstract

This paper focuses on the behaviour of cooperative, connected and automated vehicles (CCAVs) with the aim of improving traffic flow and safety and providing adequate comfort to vehicle occupants. The study is part of the AI@Edge project, funded by the Horizon Europe framework programme, which leverages AI and Edge computing to enhance 5G networks. The simulation environment is a single-lane mini-roundabout, calibrated on the basis of experimental measurements to accurately replicate the behaviour of human-driven vehicles. A cooperative Deep Reinforcement Learning (DRL) policy, exploiting Proximal Policy Optimization (PPO), was developed to optimize the behaviour of CCAVs while negotiating the roundabout. To assess the effectiveness of this policy, a dynamic driving simulator coupled with a microscopic traffic simulator and a graphical simulator was employed. This comprehensive approach included both simulated human-driven vehicles (HDs) and CCAVs, alongside a real human driver. Tests indicate that human drivers respond positively to scenarios with a higher percentage of automated vehicles, due to an enhanced sense of safety and comfort. Quantitative analysis of the policy also demonstrates the capability of CCAVs to reduce fuel consumption and optimize traffic flow.

Introduction

Urban intersections, particularly roundabouts, represent significant hurdles for the deployment of autonomous driving systems. Unlike signalized junctions with binary "stop" and "go" states, roundabouts require continuous, non-verbal negotiation between participants. This complexity is heightened in mixed-traffic scenarios where human-driven vehicles (HDs) and Cooperative, Connected, and Automated Vehicles (CCAVs) must coexist. This research explores how high-speed 5G connectivity and Edge computing can facilitate AI-driven coordination to improve traffic throughput, environmental sustainability, and passenger comfort.

Simulation Methodology and Digital Twin

To evaluate the proposed system, the researchers constructed a high-fidelity "Digital Twin" of a real-world single-lane mini-roundabout located in Milan, Italy. The simulation architecture integrated three primary components to create a realistic testing environment:

  1. SUMO (Simulation of Urban MObility): A microscopic traffic simulator used to model vehicle interactions and traffic dynamics.
  2. VI-WorldSim: A high-end graphical simulator providing a realistic visual environment for human-in-the-loop testing.
  3. DiM400 Dynamic Driving Simulator: A state-of-the-art cable-driven simulator that allowed real human participants to experience the traffic scenarios with physical motion feedback.

The simulation was meticulously calibrated using experimental data from the actual Milan site to ensure that the simulated human-driven vehicles (using the Intelligent Driver Model) accurately reflected local driving behavior.

Comparison of the SUMO network (a), VI-WorldSim graphical environment (b), and the actual roundabout in Milan (c).

AI Policy and Communication Infrastructure

The core of the CCAV control system is a Deep Reinforcement Learning (DRL) policy trained using Proximal Policy Optimization (PPO). The reward function for the DRL agent was multi-objective, aiming to maximize traffic throughput and fuel efficiency while minimizing "jerk" (the rate of change of acceleration) to ensure passenger comfort.

Communication was facilitated by a Vehicle-to-Network-to-Vehicle (V2N2V) protocol. Vehicles transmitted their state data to a Multi-access Edge Computing (MEC) server via a 5G network. The MEC server processed the AI policy and returned control instructions with the ultra-low latency required for real-time traffic negotiation, effectively moving the "brain" of the vehicle to the network edge.

Results and Discussion

The study evaluated scenarios with varying levels of CCAV penetration, ranging from 0% to 80% of the total traffic volume.

Quantitative Performance

As CCAV penetration increased, significant improvements were observed across all key performance indicators. At 80% penetration, the average crossing time for all vehicles decreased by 12.88%, and the total throughput increased from 35 to 41 vehicles per 100-second interval. Notably, the fuel consumption of CCAVs dropped significantly as they coordinated movements more efficiently. Interestingly, even human-driven vehicles saw a reduction in emissions, suggesting an "efficiency halo" where automated vehicles smooth the flow for everyone.

Qualitative Human Experience

Human-in-the-loop tests involved a panel of 10 participants navigating the simulated roundabout. Contrary to expectations that automated traffic might cause anxiety, 80% of participants rated the high-CCAV scenario as the safest. Approximately 70% of drivers explicitly preferred driving in environments with high automation, citing an enhanced sense of predictability and comfort compared to human-only traffic.

Conclusion

This research demonstrates that the integration of 5G, Edge computing, and DRL-based AI can effectively manage complex traffic nodes like roundabouts. The findings suggest that CCAVs not only optimize technical metrics like throughput and fuel efficiency but also positively influence the psychological experience of human drivers. Future research will focus on scaling these models to multi-lane intersections and utilizing physiological sensors to provide objective data on driver stress levels.

Summary updated

Abstract

Abstract The paper focuses on the behaviour of cooperative, connected and automated vehicles (CCAVs) with the aim of improving traffic flow and safety and providing adequate comfort to vehicle occupants. The study is part of AI@Edge project, founded by the Horizon Europe framework programme. AI@Edge focuses on leveraging AI and Edge computing to enhance 5G networks. The simulation environment is a single-lane mini-roundabout, calibrated on the basis of experimental measurements to accurately replicate the behaviour of human-driven vehicles. A cooperative Deep Reinforcement Learning (DRL) policy, exploiting Proximal Policy Optimization (PPO), was developed to optimize the behaviour of CCAVs while negotiating the roundabout. To assess the effectiveness of this policy, a dynamic driving simulator coupled with a microscopic traffic simulator and a graphical simulator was employed. This comprehensive approach included both simulated human-driven vehicles (HDs) and CCAVs, alongside a real human driver. Tests indicate that human drivers respond positively to scenarios with a higher percentage of automated vehicles, due to an enhanced sense of safety and comfort. Quantitative analysis of the policy also demonstrates the capability of CCAVs to reduce fuel consumption and optimize traffic flow.
BibTeX
@inproceedings{previati2025future,
  author = {Giorgio Previati and Elena Campi and Lorenzo Uccello and Antonino Albanese and Alessandro Roccasalva and Gabriele Santin and Massimiliano Luca and Bruno Lepri and Laura Ferrarotti and Nicola di Pietro and Marco Ponti and Gianpiero Mastinu},
  title = {{Future Roundabouts Relying on 5G, Edge Computing and Artificial Intelligence}},
  booktitle = {Lecture Notes in Mobility},
  year = {2025},
  doi = {10.1007/978-3-032-06763-0\_77},
}
Figure