Distributed Artificial Intelligence for Traffic Assignment in Smart Cities

This paper aims to contribute to the challenging issue of one microscopic simulation round for dynamic traffic assignment. It relays on the selfish behaviour of the vehicle agent that benefits from a more accurate estimation of its travel time. The main novelty is that, rather than considering the average travel times in the network links according to the present vehicles, the vehicle must first know when it can cross the nodes located at both extremities of the road. This is achieved through a negotiation between the vehicle agent and the node agents to book the crossing time. In order to assess this new paradigm, this paper compares it with well-known approaches in an elementary network. The result invites us to extend the approach to more general cases. A discussion of the opportunities and limitations of the approach extension is provided in this paper. One of the notable opportunities is that the proposed approach has a great potential to improve energy efficiency by exploring mobile navigation applications and connected autonomous vehicles.