JACK Intelligent Agents – Summary of anAgent InfrastructureNick Howden, Ralph Rönnquist, Andrew Hodgson, Andrew LucasAgent Oriented Software Pty. Ltd.221 Bouverie St, CarltonVictoria 3053, Australia+61 3 9349 5055http://www.agent-software.com<firstname.surname>@agent-software.comABSTRACTIntelligent Agents are being used for modelling simple rationalbehaviours in a wide range of distributed applications. Inparticular, multi-agent architectures based on the Belief-Desire-Intention model have been used successfully in situations wheremodelling of human reasoning and team behaviour are needed,such as simulating tactical decision-making in defence operationsand command and control structures. Other applications includeintelligent web applications, manufacturing control, telephone callcentres, and air traffic management.The JACK Intelligent Agents framework by Agent OrientedSoftware brings the concept of intelligent agents into themainstream of commercial software engineering and Java. JACKIntelligent Agents™ is a third generation agent framework,designed as a set of lightweight components with highperformance and strong data typing.We present the design approach and major technicalcharacteristics of JACK Intelligent Agents, with a focus onsome of the more recent developments – modular structuringelements called Capabilities and the team-based reasoning modelcalled SimpleTeam. Two example applications, in decisionsupport and defence simulation are described. Also, we discussthe benefits of the component-based approach, both for thesoftware engineer developing sophisticated distributedapplications, and for the researcher exploring agent models andarchitectures.KeywordsIntelligent Agents, Infrastructure, JACK, BDI, Java, Multi-agentSystems.1. INTRODUCTIONIntelligent Agents are being used for modelling simple rationalbehaviours in a wide range of distributed applications. Intelligentagents have received various, if not contradictory, definitions; bygeneral consensus, they must show some degree of autonomy,social ability, and combine pro-active and reactive behaviour [1].One of the better known and most successful architectures foragents is the so-called BDI (Belief-Desire-Intention) architecture,which has seen a number of academic and industrial applications.Agent Oriented Software Pty. Ltd. (AOS), based in Melbourne,Australia, has built JACK Intelligent Agents, a framework inJava for multi-agent system development. The company’s aim isto provide a platform for commercial, industrial and researchapplications. To this end, its framework supplies a highperformance, lightweight implementation of the BDI architecture,and can be easily extended to support different agent models orspecific application requirements. For brevity, we will refer toJACK Intelligent Agents simply as ‘JACK’.This paper is organised as follows. Section 2 introduces JACKIntelligent Agents™, presenting the approach taken by AOS to itsdesign, and outlining its major engineering characteristics. TheBDI model is discussed briefly in Section 3. Sections 4 & 5describe some of the more interesting components of the JACKinfrastructure and Section 6 gives some example applications builtwith JACK Intelligent Agents™. The JACK components andtools and the future research and development direction isdescribed in Sections 7 & 8. Finally, in Section 9 we discuss howthe use of this framework can be beneficial to both engineers andresearchers.2. JACK Intelligent AgentsIn this section, we present JACK by first highlighting the goals setby its designers, then we provide an overview of the engineeringcharacteristics of the framework.2.1 ApproachMajor design goals for JACK were: to provide developers with a robust, stable, light-weightproduct; to satisfy a variety of practical application needs; to ease technology transfer from research to industry; and to enable further applied research.Whilst applications can be built from the ground up adopting anagent oriented methodology and an appropriate framework, mostorganisations already possess and depend upon large legacysoftware systems. Thus, JACK agents have been designedspecifically for use as components of larger environments.Consequently, an agent must coexist and be visible as simplyanother object by non-agent software. Conversely, a JACKprogrammer must be allowed to easily access any othercomponent of a system. Type safeness when accessing data,reliability, and support for proper engineering processes are thenkey requirements in this kind of environment.For similar reasons, JACK agents are not bound to any specificagent communications language. Nothing prevents the adoptionof high-level symbolic protocol such as KQML or FIPA’s AgentCommunication Language (ACL); possibly by integratingsoftware already existing in the public domain. However, JACKhas been geared towards industrial object-oriented middleware(such as CORBA) and message passing infrastructures (forinstance, HLA or DIS in simulation environments). In addition,JACK provides a native lightweight communicationsinfrastructure for situations where high performance is required.JACK itself has been designed for extension by properly trainedengineers, familiar with agent concepts and with a soundunderstanding of concurrent object-oriented programming.2.2 Overview of the frameworkFrom an engineering perspective, JACK consists of architecture-independent facilities, plus a set of plug-in components thataddress the requirements of specific agent architectures. The plug-ins supplied with version 3.0, released in February 2001, includesupport for the BDI model, and the model for building teams ofagents called SimpleTeam.To an application programmer, JACK currently consists of threemain extensions to Java. The first is a set of syntactical additionsto its host language. These additions, in turn, can be divided asfollows: a small number of keywords for the identification of the maincomponents of an agent (such as agent, plan and event); a set of statements for the declaration of attributes and othercharacteristics of the components (for instance, theinformation contained in beliefs or carried by events). Allattributes are strongly typed; a set of statements for the definition of static relationships(for instance, which plans can be adopted to react to a certainevent); a set of statements for the manipulation of an agent's state(for instance, additions of new goals or