ON APPLICATIONS AND ENVIRONMENTS FOR MULTI-AGENT SYSTEMS Paul Valckenaersa John Sauterb Carles Sierrac Juan Antonio Rodriguezc a Department of Mechanical Engineering, Katholieke Universiteit Leuven, Belgium, paul.valckenaers@mech.kuleuven.be, b NewVectors, LLC, USA c The Artificial Intelligence Research Institute of the Spanish Research Council, Spain Abstract This paper addresses MAS environments from an application perspective. It presents a structured view on environment-centric MAS applications. This comprises three base configurations, which MAS applications may apply directly or combine into a composite configuration. For each configuration, the paper presents key issues, requirements and opportunities (e.g. time management issues, real-world augmentation opportunities and state snapshot requirements). Thus, the paper delineates what environment technology may implement to serve MAS applications. Sample applications illustrate the configurations. Electronic institutions provide an example of an environment technology, already achieving some maturity. This paper is a preview of a more extensive paper (Autonomous Agents and Multi-Agent Systems (2007) 14:61–85, DOI 10.1007/s10458-006-9002-5). 1 Simulation Applications In simulation applications, agents interact with the environment, which models entities in the world-of- interest. There is no run-time connection from the real world to these models in the environment. In this area, repositories of reusable emulation models can capture important domain knowledge and improve the overall quality of the models employed in MAS. The fact that these emulation models correspond to parts Progress of computations and time of the real world facilitates integration Agent system and reusability significantly. To this end, Agent emulation model developers must avoid Agent relying on a specific context and only Time management Agent Agent reflect the corresponding part of the world-of-interest. Specific for the simulation context is Environment the need for time management Virtual world: real world emulation functions. Some of these functions are the subject of ongoing research. In Environment Environment entity entity contrast, the ability to slow down agent Environment computations when the emulation entity Environment entity Simulated time requires more than real-time, or the ability to emulate in real-time while agents are deliberating in combination with the speed-up of discrete-event simulation Up Front off-line modeling when agents are idling, are relatively simple time management functions. World-of-interest Nonetheless, such functionality is absent Real World Actor in existing simulation software and cannot Environment Entity (deployed) be added later. Environment technology has the opportunity to provide suitable Entity Entity Entity support at the core of its implementations. Entity Entity 2 Real-world Interaction A second configuration targets applications rooted in the physical world. Agents handle all decision- making aspects; the environment reflects the remaining entities in the world-of-interest. There is a run- time connection between the environment and the real world. The agents only interact with the real world through the environment as an intermediary. In this configuration, the environment provides important augmentations of the real world. It provides enhanced access to the Agent system Agent real world (sensor data processing, virtual sensors and actuators), on- Agent line auto-updated documentation Agent Agent (specs, maps), and information processing infrastructures (stig- mergy). In additon, the environ- ment regulates the usage of and Environment access to real-world entities. Virtual world: real world sensing, Moreover, it extends the real- acting and emulating world by its past (log or trace) and Environment entity its future (what-if models, Environment possibly reflecting intentions and entity Real time commitments). Environment and possibly Environment entity historical log, This functionality typically is entity future behavior reusable and modular. Indeed, many of the environment entities can be developed based on self- World-of-interest knowledge (of the corresponding real-world entity) only. Hence, Real World such augmented environment entities can be constructed to be Entity reusable wherever and whenever Entity an instance of the correspond-ing Entity Entity Entity real-world entity exists. And, much of the functionality is application-independent. 3 Other configurations In a third configuration, the environment augments adaptive structured information systems. These augmentations extend the information system, monitor the information system and add information processing structures. Furthermore, the environment captures the behavior of the agents themselves, allowing the agents to benefit from past experience of the agent community. Moreover, the environment may provide links between information structures. Composite configurations reveal how base configurations must anticipate integration requirements. The functionality identified in the composite configuration includes access to the ‘system state’ in base configurations. Time and time management services are also important. Finally, the paper addresses software support. Electronic Institutions are discussed as an example of relevant technology. Acknowledgements This paper presents work funded by the Research Fund of the K.U.Leuven (Concerted Research Action on Autonomic Computing for Distributed Production Systems), and the projects Web-i(2) and OpenKnowledge.