Multiagent architectures for intelligent traffic management systems

Abstract

This paper reports our experiences with agent-based architectures for intelligent traffic management systems. We describe and compare integrated TRYS and TRYS autonomous agents, two multiagent systems that perform decision support for real-time traffic management in the urban motorway network around Barcelona. Both systems draw upon traffic management agents that use similar knowledge-based reasoning techniques in order to deal with local traffic problems. Still, the former achieves agent coordination based on a traditional centralized mechanism, while in the latter coordination emerges upon the lateral interaction of autonomous traffic management agents. We evaluate the potentials and drawbacks of both multiagent architectures for the domain, and develop some conclusions respecting the general applicability of multiagent architectures for intelligent traffic management.

Introduction

The notion of software agents has become increasingly popular over the last couple of years. Public entities as well as private companies spend a considerable amount of time, effort and money in research development and promotion of the idea of software agents. To some respect this is surprising, as software agents essentially do what one expects a reasonable advanced computer program: it is embedded in an environment in which it is capable of achieving certain tasks with some degree of autonomy, i.e. without constant human guidance or intervention.

Still, the real potential of this technology becomes unleashed when several software agents are put to use in the same environment. In this case, the group of agents is usually conceived as a multiagent system, as the successful completion of their tasks is subject to the decisions and actions of other agents. So, in multiagent systems, agents are forced to coordinate their activities so as to avoid negative interactions with their acquaintances and to exploit synergic potentials. Initially, most distributed problem-solving systems were based on a distinguished agent, which achieved the coordination of the activities of its acquaintances in a centralized fashion (Steeb et al., 1988; Von Martial, 1992). More recently, the focus has shifted to more autonomous agents that coordinate in a decentralized fashion (Ephrati et al., 1995; Decker and Lesser, 1998).

An important reason for the growing success of multiagent technology is its potential to cope with high complexity problems that show an a priori distribution (Ossowski, 1999). From the technical point of view, the inherent distribution allows for a natural decomposition of the system into agents that interact so as to achieve a desired global functionality. By this, reusability is promoted, not just by the agents’ modularity, but also based on their autonomy. In addition, the scalability of systems operating in highly complex domains can be improved by choosing among specific coordination models that harmonize agent activities with respect to the multiagent system’s task. Through an adequate design of coordination mechanisms for a particular problem, it is believed that, from an economic point of view, multiagent systems may tackle complex problem with an acceptable degree of performance, but at a lower cost than traditional solutions.

Multiagent systems have been applied successfully to a variety of industrial problems (Chaib-Draa, 1998), from electronic commerce (Guttman et al., 1998; Foss, 1998) through energy management (Jennings, 1994; Correra et al., 1994), to road transportation related applications such as parking guidance tools or transportation scheduling systems (Fischer et al., 1996). So, it is natural to ask whether this promising new technology is capable of coping efficiently with advanced road traffic management scenarios. In the end, traffic management problems are intrinsically distributed, and suffer from a high degree of complexity that is constantly increasing in line with the incessant improvements of traffic control infrastructures.

This article reports our research and experiences in this respect: it analyzes how multiagent architectures can be applied to the problem of strategic road traffic management. Section 2 discusses the role of knowledge-based artificial intelligence techniques for traffic management. It argues in favor of the concept of intelligent traffic management systems (ITMS) as a means of integrating the increasingly complex and heterogeneous traffic control infrastructure and providing a means of strategic support for traffic management. Section 3 shows how knowledge-based ITMS for real-world traffic management problems can be instrumented in a computationally and economically efficient way by means of multiagent architectures. Two ITMS aimed at the management of the Barcelona freeway system, and based on different multiagent architectures, are described and compared. Finally, in Section 4, the potentials and drawbacks of both systems are analyzed in order to come up with a set of conclusions respecting the applicability of multiagent technology to the ITMS domain.