One of the major challenges in engineering distributed multiagent systems is the coordination necessary to align the behavior of different agents. Decentralization of control implies a style of coordination in which the agents cooperate as peers with respect to each other and no agent has global control over the system, or global knowledge about the system. The dynamic interactions and collaborations among agents are usually structured and managed by means of roles and organizations. In existing approaches agents typically have a dual responsibility: on the one hand playing roles within the organization, on the other hand managing the life-cycle of the organization itself, for example, setting up the organization and managing organization dynamics. Engineering realistic multiagent systems in which agents encapsulate this dual responsibility is a complex task. In this article, we present a middleware for context-driven dynamic agent organizations. The middleware is part of an integrated approach, called MACODO: Middleware Architecture for COntext-driven Dynamic agent Organizations. The complementary part of the MACODO approach is an organization model that defines abstractions to support application developers in describing dynamic organizations, as described in Weyns et al. [2010]. The MACODO middleware offers the life-cycle management of dynamic organizations as a reusable service separated from the agents, which makes it easier to understand, design, and manage dynamic organizations in multiagent systems. We give a detailed description of the software architecture of the MADOCO middleware. The software architecture describes the essential building blocks of a distributed middleware platform that supports the MACODO organization model. We used the middleware architecture to develop a prototype middleware platform for a traffic monitoring application. We evaluate the MACODO middeware architecture by assessing the adaptability, scalability, and robustness of the prototype platform.

Real environments in which agents operate are inherently dynamic -the environment changes beyond the agents' control. We advocate that, for multi-agent simulation, this dynamism must be modeled explicitly as part of the simulated environment, preferably using concepts and constructs that relate to the real world. In this paper, we describe such concepts and constructs, and we provide a formal framework to unambiguously specify their relations and meaning. We apply the formal framework to model a dynamic RoboCup Soccer environment and elaborate on how the framework poses new challenges for exploring the modeling of environments in multi-agent simulation.

This paper presents an agent-based approach, facilities. This enables interesting applications such as vecalled delegate multi-agent systems, for anticipatory vehicle hicles that collaboratively interpret the local traffic situation routing to avoid traffic congestion. In this approach, individual and spread useful information to traffic signs that inform vehicles are represented by agents, which themselves issue lightweight agents that explore alternative routes in the environment drivers about the actual traffic conditions. In this paper, on behalf of the vehicles. Based on the evaluation of the we focus on another interesting application: anticipatory alternatives, the vehicles then issue light-weight agents for vehicle routing to avoid traffic congestion. Congestion is allocating road segments, spreading the vehicles' intentions and basically a resource coordination problem where vehicles coordinating their behavior. To evaluate the approach, we have have conflicting intentions about the use of parts of the road developed an initial prototype application. Test results indicate that delegate multi-agent systems are a promising approach for network One way to avoid (or at least reduce) congestion iS anticipatory vehicle routing.by letting vehicles anticipate possible conflicts. In this paper, we propose a MAS-based approach called "delegate MAS" I. INTRODUCTION for anticipatory vehicle routing. In this approach, individualMonitor and control systems for traffic share two im-vehicles issue light-weight agents that explore alternative portant characteristics with other complex distributed soft-paths in the environment on behalf of the vehicles. Based ware systems: (1) highly dynamic and changing operation on the evaluation of the alternatives, the vehicles then issue conditions under which the systems have to operate such a second type of light-weight agents for allocating road as a fluctuating amount of traffic, changing behavior of segments, spreading the vehicles' intentions and coordinating drivers, traffic jams, and road accidents, and (2) the inherent their behavior. For experimentally evaluating the approach, distribution of resources and activity making centralized we have developed an initial prototype application. The work control hard to achieve; for example: traffic is naturally we present is obviously still in progress, and it is clear distributed over the road network, traffic lights act locally (or that there is still a lot of research to be fulfilled (both are coordinated regionally), traffic jams cause local delays, conceptually as well as technologically) before we can assess etc. Examples of other classes of systems that share these the feasibility of the approach in practice. However, test characteristics are manufacturing control systems, automated results show that the approach is quite promising.transportation systems, and wireless sensor networks.Overview. The remainder of this paper is structured asIn our research, we study situated multi-agent systems follows. In section II, we elaborate...

Today's distributed applications such as sensor networks, mobile multimedia applications, and intelligent transportation systems pose huge engineering challenges. Such systems often comprise different components that interact with each other as peers, as such forming a decentralized system. The system components and collaborations change over time, often in unanticipated ways. Multiagent systems belong to a class of decentralized systems that are known for realizing qualities such as adaptability, robustness, and scalability in such environments. A typical way to structure and manage interactions among agents is by means of organizations. Existing approaches usually endow agents with a dual responsibility: on the one hand agents have to play roles providing the associated functionality in the organization, on the other hand agents are responsible for setting up organizations and managing organization dynamics. Engineering realistic multiagent systems in which agents encapsulate this dual responsibility is a complex task.In this article, we present an organization model for context-driven dynamic agent organizations. The model defines abstractions that support application developers to describe dynamic organizations. The organization model is part of an integrated approach, called MACODO: Middleware Architecture for COntext-driven Dynamic agent Organizations. The complementary part of the MACODO approach is a middleware platform that supports the distributed execution of dynamic organizations specified using the abstractions, as described in Weyns et al. [2009].In the model, the life-cycle management of dynamic organizations is separated from the agents: organizations are first-class citizens, and their dynamics are governed by laws. The laws specify how changes in the system (e.g., an agent joins an organization) and changes in the context (e.g., information observed in the environment) lead to dynamic reorganizations. As such, the model makes it easier to understand and specify dynamic organizations in multiagent systems, and promotes reusing the life-cycle management of dynamic organizations. The organization model is formally described to specify the semantics of the abstractions, and ensure its type safety. We apply the organization model to specify dynamic organizations for a traffic monitoring application. ACM Reference Format:Weyns, D., Haesevoets, R., and Helleboogh, A. 2010. The MACODO organization model for contextdriven dynamic agent organizations.

Interaction is at the core of multi-agent systems. We use agent environment as a general term to denote the medium for agent interaction. Over the last years, the agent environment has been subject of active research. In this paper, we reflect on the role of the agent environment in multi-agent systems from a middleware perspective. Our study yields the following observations: (1) multi-agent system engineers consider distributed middleware (RMI, CORBA, etc.) as the basic platform for developing multi-agent systems, (2) common middleware services (security, persistency, etc.) are only minimally considered in multi-agent systems, (3) domain-specific middleware for multi-agent systems such as communication services and support for stigmergic coordination are typically developed as stand-alone services and as such difficult to compose with other services. From these observations, we derive a number of challenges for research on environments in multi-agent systems: (1) to amplify reuse, application-specific services should be further consolidated into domain-specific services, (2) the problem of integration must be tackled, i.e. horizontal integration among domain-specific services for multi-agent systems, and vertical integration of domain-specific services upwards with the agents, and downwards with the common middleware services and the underlying distributed platform, (3) to support dynamic changing requirements of the system at hand, flexible composition and dynamic adaptation of services must be supported by the agent environment.