Technical Report 2

The Potential for Intelligent Software Agents in Defence Simulation

by Andrew Lucas and Simon Goss

Introduction Only

Our objective is to describe the application of intelligent software agents to military simulation. simulation is used to support procurement, force development, evaluation of C3 structures and for training. In all these applications it is necessary to model both individual human reasoning and team We describe the intelligent agent and outline the current state of the art in agent technology, including the of team and agents that can learn. We cover a number of recent and current intelligent agent applications that are demonstrating the potential for agents to effectively represent human reasoning and teams in C3 models.

The term agent is widely used to describe a range of software varying in capability from the procedural wizards found in popular desktop applications, to information agents for information search and retrieval, and to intelligent agents capable of simple rational reasoning. The intelligent agent as described here is an autonomous piece of software, which has explicit goals or desires to achieve, and is preprogrammed with plans or to achieve these goals under varying circumstances. Set to work, the agent pursues its given goals adopting the appropriate plans, or intentions, according to its current beliefs about the state of the world, so as to perform the role it has been given. Such an intelligent agent is generally referred to as a Belief-Desire-Intention (BDI) agent.

Under the BDI model, agents may be given pre-compiled or they may plan or learn new plans at execution time. Giving BDI agents pre-compiled plans is a method for ensuring predictable under critical operational conditions, and for ensuring performance.

BDI agents are highly suited to the development of time and mission critical systems, as the BDI approach provides for the verification and validation of the model. The agents goals may include keeping the human users informed of what the agent is trying to achieve, what its current intentions are, and how far it has got.

The ability of intelligent agents to perform simple tasks autonomously has aroused much interest in the potential military applications. Key characteristics of intelligent agents that make them attractive are:

• autonomy;
• high-level representation of - easy to define command and control architectures;
• flexible combination of proactivity and reactivity;
• real-time performance;
• suitability for distributed applications; and
• ability to work cooperatively in teams.

The development of intelligent agents has evolved from the early Artificial Intelligence research into the development of autonomous mission critical software technologies. Initial concepts for intelligent agents were explored at SRI International by Georgeff and Lansky (1986) in the mid-1980s and later by Rao and Georgeff (1990) in the early 1990s. An early implementation was the development of the LISP-based Procedural Reasoning System (PRS). The SOAR system was also developed in the USA at this time and has since been used by ISI at the University of Southern California for prototype applications.

Research into distributed real-time AI systems and agent architectures at the Australian Artificial Intelligence Institute (AAII) in the early 1990s by Rao, Georgeff and others (Wooldridge & Rao, 1997) resulted in the development of second-generation dMARS C++ multi-agent system. Parallel development by Ingrand in France led to the C-PRS single agent system (Ingrand et al, 1996).

Current developments include two JAVA-based agent developments: the BT Laboratories, UK ZEUS agents toolkit (Nwana et al, 1998); and the JAM system from IRS, USA. engineers and researchers.


back