Smart multimodal corridor control: integrated AI and model-based strategies for urban and extra-urban traffic management

This thesis develops an integrated framework for smart multimodal corridor control, combining model-based and AI-driven strategies to enhance the efficiency, equity, and sustainability of urban and extra-urban traffic management systems. The research contributes to the activities of the MOST – National Center for Sustainable Mobility (Spoke 9), addressing coordinated control of public and private transport under both deterministic and stochastic conditions. In the urban context, two complementary approaches are investigated at the intersection level. The first introduces a facility-based bus priority design through pre- signaling, where a dedicated bus lane terminates upstream of the intersection and is managed via an auxiliary pre-signal. Through microsimulation in SUMO, different configurations of pre-signal placement and control logic are analyzed along a real corridor in Rome. Results demonstrate that appropriately positioned pre-signals significantly reduce delay for buses, especially for left-turn movements, while maintaining acceptable levels of general traffic performance. The second urban approach applies Deep Reinforcement Learning (DRL) to adaptive signal control and Transit Signal Priority (TSP) in a large-scale real network, the Via Prenestina corridor, including 21 signalized intersections. The DRL agent integrates a convolutional neural network for traffic state representation and a reward structure that balances private and public transport performance. Simulation outcomes confirm that the proposed DRL-based system reduces total delay, queue lengths, and emissions while improving the regularity of public transport service compared to both fixed-time and traditional adaptive controls. In the urban freeway context, the thesis introduces a dynamic rule-based Variable Speed Limit (VSL) control model, operationally validated on the Padua–Mestre motorway. The model accounts for driver behavior and stochastic compliance, using field data 2 from nearly one year of real observations. Statistical testing and clustering analysis show that the dynamic VSL reduces travel time by up to 19% and decreases speed variance by more than 50% under congested conditions. To test the applicability of our dynamic rule-based model, we extended our analysis using SUMO microsimulation to a new study area on the Napoli ring road, using a stochastic user-compliance simulation model in which drivers are categorized as compliant or non-compliant based on a long-normal distribution of the speed factor. Finally, we compared the rule-based strategy with a DRL-based jam-shrinking control, highlighting the practicality of the rule-based solution for deployment. Overall, the thesis delivers a coherent methodological and experimental contribution that not only bridges model-based, rule-based, and AI-based adaptive control paradigms but also tests various strategies in a real-world simulated environment. We conclude that fair and practical comparisons between these approaches are essential for guiding future implementations of intelligent and cooperative traffic management within sustainable mobility systems.