Operational Knowledge Representation - Behaviour Capture, Modelling and Verification Nick Howden; Jamie Curmi Agent Oriented Software Pty Ltd <firstname.surname>@agent-software.com Clinton Heinze; Dr Simon Goss Air Operations Division, Defence Science and Technology Organisation <firstname.surname>@dsto.defence.gov.au WGCDR Grant Murphy Australian Defence Simulation Office grant.murphy@cbr.defence.gov.au Abstract. In simulation, behaviour modelling traditionally takes a back seat to physical representation and realistic 3D graphics. Recently this has started to change -- most notably, agent technologies have shown themselves to be a powerful tool for modelling complex decision-making. Agents are increasingly used in simulation where they prove highly effective, both for training and analysis. However, when building and running an agent-based simulation, agents must be more accessible to human analysts and subject matter experts. It is vital that the agent reasoning processes are visible, allowing verification of agent behaviours without the constant need for programming expertise. These requirements led to the establishment of the Operational Knowledge Representation Aid (OKRA) project. Developed by Agent Oriented Software (AOS), this project was initiated and funded by the Australian Defence Simulation Office (ADSO) with technical and management assistance from Air Operations Division (AOD) of DSTO. This paper outlines practical mechanisms for supporting the development of intelligent behaviour in simulation. It examines graphical methods for the design and definition of human behaviours as agent plans, and tracing tools that display behaviours and decision-making of agents in action. Together, tools in the OKRA suite facilitate the capture of knowledge from operational personnel and subject matter experts, along with the verification of that knowledge in a running system. Users can outline behaviours in natural language, then convert these outlines into code that implements those behaviours. 1. INTRODUCTION The Australian Defence Simulation Policy describes a future where simulation is ubiquitous and coherent across Defence, for training, planning and decision support in all operational and departmental processes. A significant pre-requisite to the development of simulation is the development of more sophisticated representations of human behaviour than have previously been available. For some time Defence has been researching Intelligent Agent technology within DSTO, specifically using the Beliefs-Desires-Intentions (BDI) model and the JACK Intelligent Agents™ (JACK) product from Agent Oriented Software, as a way of advancing the representation of human behaviour. These activities have highlighted the potential, for example, for Agents to: • increase the complexity of issues which can be replicated within a simulation; • increase the richness of experience to which trainees are exposed; • facilitate large scale collective training with simulated intelligent adversaries and allies; and • reduce the cost and hence increase the attraction of simulation, by reducing the numbers of human ‘players’ required. Defence’s experience with Agents revealed two impediments to larger-scale uptake of the technology. The first was a programming overhead. Human analysts and operational specialists (typically members of the ADF) defined the required Agent behaviour. However, specialised and scarce programmers were required to assign this behaviour to a specific Agent within the simulation system. The second impediment concerned confidence: the intermediate programming activity between the analyst/specialist and the Agent complicated the essential process of confirming that the Agent properly exhibited the intended behaviour. ADSO has a broad responsibility for advancing simulation in Defence. Confidence building and technology development are therefore issues of considerable interest to ADSO. Consequently, ADSO sponsored the development of a suite of graphical tools to allow analysts to assign Agent behaviour while minimising the need for intermediate programming, and to allow Agent behaviour to be monitored by operational specialists during execution. The interface