The transport sector is expected to alter shape in the coming decades. This transition is driven by user-centred mobility services, integrated and intelligent transport networks, automation, as well as public and private innovation. Next to Big Data and Artificial Intelligence, these developments provide major opportunities, but also increase system complexity. The overarching aim of DIT4TraM is to develop a generic distributed control paradigm, applicable at the level of traffic operations, mobility management, demand-supply synchronization and shared mobility, including advanced monitoring and machine learning technology for a variety of novel multi-modal management and mobility concepts operating at all urban scales. A holistic approach to decentralization, distribution and mechanism design for monitoring and control is proposed aiming to achieve social optimality. Our vision is to support the transition to seamless and sustainable connected and autonomous mobility by disentangling the system components to the highest extent possible, yet ensuring sufficient cooperation and emergent coordination by smart system design, leading to livability, safety, resilience, efficiency, as well as privacy, participation, fairness and sustainability on a city scale.
How We Contribute
The DSAIT team contributed to the DIT4TraM project by developing and testing distributed traffic management strategies for future urban mobility systems. The focus was on decentralized approaches for perimeter control, dynamic space allocation and auction-based traffic signal control to improve traffic flow, reduce congestion and support multimodal transport through adaptive real-time decision making. Leveraging real traffic data from Athens and complex microsimulation environments, we demonstrated how Reinforcement Learning and Imitation Learning could be combined to develop efficient model-free and distributed perimeter control mechanisms at different decentralization levels to maximize the travel production of the urban network with minimal central oversight. In addition, we developed a range of dynamic space allocation mechanisms at both the intersection and corridor levels. One category focused on mode-based allocation strategies, such as dynamic bus lanes and priority lane density control. The other category targeted reconfigurable network topologies, specifically through the use of dynamic reversible lanes. These mechanisms were designed and optimized using reinforcement learning techniques. Finally, as part of our innovative contributions, we also explored auction-based traffic signal control under sealed-bid second-price auctions combined with a delay-based bid-reservation allocation mechanism. This approach aims to achieve fair and efficient allocation of green time among competing traffic streams and thus avoiding starvation phenomena. Collectively, these methods formed a core component of the project’s vision for future intelligent transport systems.
