Pii: S0921-8890(99)00006-8

In the late seventies and early eighties, multi-agent systems (MASs) emerged as a scientific area from several previous research domains, such as the distributed artificial intelhgence, intelligent robotics, and knowledge-based Systems. This area of research is now reaching a level cf maturity, enough for MAS to be applied as a technology for solving problems in complex applications. Using MAS technology has proven to properly support two wide, distinct, and generic classes of complex application environments. One such class represents the domain of applications that, by nature, are characterized as physically ditstributed networks of nodes requiting some level of autonomy. This area comprises a very large domain of applications. Examples of such applications include the domains of computer integrated manufacturing (CIM), networked organizations such as the supply chains and hospitals, and the virtual enterprises for manufacturing industry, tourism, and transportation, among others. For example in CIM, upto-date manufacturing methods are based on efficient evaluation of vast volumes of (usually distributed) data and knowledge of diverse nature, including the top-level expertise knowledge. These methods rely on distributed problem solving and advanced control algorithms exploring distributed resources. Virtual and chain organizations similarly require the support for inter-operability and task coordination among heterogeneous and autonomous systems. The MAS technology provides the necessary infrastructure for this entire area. The other class represents the wide variety of applications that even if by nature are centralized, can vastly benefit from a logically distributed software implementation approach. In connection with the rapidly growing capabilities of computers, requirements for software solutions are becoming more and more complex. Any monolithic software solution cannot cover the complex and very often vaguely specified and contradictory requirements set for the information management and decision-making systems. Among others, these requirements usually include quite strict needs for the quality assurance, and for the simplicity of software maintenance. Furthermore, flexibility in terms of global interoperation of functionalities is becoming an important requirement for agile organizations. To support such a criteria, the complex software solutions are usually constructed as integration of smaller, well-specified, well-approved, and comparatively compact software units. Flexible, intelligent communication and cooperation of these units seem to be a crucial aspect of the creation of composed systems. Cooperative multi-agent systems offer a novel approach to handle these complex systems. Unlike more traditional modularity methods of software engineering, the MAS approach can more easily cope with the heterogeneity and autonomy of participating components and therefore be more suitable for the integration of legacy systems. This integration can be done, for example, by an agentification process, where a so-called "agent wrapper" is built around each of the already well-operating software units. This wrapper contains, for instance, acquaintance models as well as the communication and negotiation functionalities, enabling the inclusion of the legacy unit into the wider multi-agent community. Application examples of MAS can be found in nearly all application domains, including the integrated enterprise management, technical and medical diagnostics, production planning and supervision, air traffic control, electronic businesses, entertainment systems, etc. But the industrial applications were