Vehicle-centric coordination for urban road traffic management: A market-based multiagent approach

Abstract

Traffic congestion in urban road networks is a costly problem that affects all major cities in developed countries. For example, the Texas Transportation Institute estimated that traffic jams in the U.S. cost more than 78 billions dollars every year, in fuel consumption and productivity loss [1]. To tackle this problem, it is possible i) to increase the capacity of the network, adding more lanes or more roads, ii) to reduce the demand, restricting the access to urban areas at specific hours or to specific vehicles, or iii) to improve the efi- ciency of the existing network, by means of a widespread use so-called Intelligent Transportation Systems [41]. In line with the recent advances in telematic infrastructures, the traffic control and management problem has turned out to be a promising application field for multiagent system technology [115]. Multiagent systems (MAS) are the ideal candidates for the design and implementation of such systems, since many problems in this domain are inherently distributed and the actors _t perfectly the paradigm of autonomous agent [11]. In this thesis, several distributed, market-based, mechanisms have been studied and applied to the management of a (future) urban road network, where intelligent autonomous vehicles, governed by driver agents, interact with the infrastructure in order to travel through the network. Starting from the reservation-based intersection model proposed by Dresner and Stone in [35], this thesis studied how to implement a computational economy where the driver agents must acquire the necessary reservations to cross the intersections that compose their routes, while the agents in charge of managing the intersections (intersection managers) participate in the market as suppliers of such reservations. Two scenarios have been studied, one with a single intersection and one with a network of intersections. In the first case, we have developed different policies to control a reservation-based intersection, based on the adversarial queueing theory and the combinatorial auction theory. In the second case, we have studied two different models of computational economy to deal with the traffic assignment problem. The first one, ECO+, is a cooperative model, where the intersection managers learn to operate in the market to optimise a global profit measure for the society of intersection managers and, indirectly, the travel time of the driver agents. The second one, ECO¿¿, is a competitive model, where the intersection managers compete with each other as suppliers of the reservations that are traded in the market, aiming at reaching the market equilibrium, that is, a situation where the amount of resources sought by buyers (driver agents) is equal to the amount of resources produced by suppliers (intersection managers). Finally, we combined the auction-based policy for traffic control and the competitive model for traffic assignment into an adaptive, integrated, strategy for full-edged traffic management, ECO¿¿ CA. In parallel to the theoretical design of the market-based mechanisms, in this thesis we developed a simulation tool, called M:I:T :E: (Multiagent Intelligent Transportation Environment), to evaluate the proposed mechanisms and to show how these mechanisms affect the driver agents' utility as well as the system utility. This simulator implements two validated traffic ow models (the mesoscopic model of Schwerdtfeger [89] and the microscopic model of Nagel and Schreckenberg [68]), and provide a powerful tool that enables the simulation of thousands of vehicles with high precision.