In military Modelling and Simulation (M&S), there is an increasing need for Computer Generated Forces (CGFs) with advanced autonomous behaviours for use in training, Concept Development & Experimentation (CD&E) and decision support. However, the development of behaviour models is often resource-intensive and reuse across applications, simulation systems or scenarios is difficult. This is clearly illustrated by the plethora of platform-specific (proprietary) modelling tools on the market and is considered a burden throughout interviews conducted with staff operating various simulators of the Dutch military.
Our paper regards the CGF behaviour modelling process from the perspective of different stakeholders such as subject-matter experts (SMEs), designers, developers and operators/instructors, with a twofold purpose. We both sketch a holistic overview of current challenges and opportunities for reusability, and we propose three synergetic directions for improving reusability.
The first focuses on technology-independent behaviour descriptions which facilitates SME knowledge elicitation and allows for (semi)-automated conversion to executable models.
The second concerns the use of machine learning techniques to generate (parts) of a behaviour model by making use of military reference material such as doctrine documents or rules of engagement.
Finally we propose a general-purpose executable behaviour construct, modelled specifically for reusability. As a reusable construct, it can represent any level of behaviour (tactics, procedures and/or actions) and is agnostic to specific implementation techniques. Being self-contained makes it suitable for repository storage and cloud execution, in line with NATO efforts on M&S resource management and M&S as a Service.
The above techniques are evaluated jointly in a use case for air-to-air combat behaviour. We demonstrate the reusability of tactical behaviours across two simulation systems, for CD&E purposes and fighter pilot training. The behaviours employ both hand-crafted and machine learning techniques and are (partly) generated from (reusable) behaviour descriptions created by former F-16 pilots acting as SMEs.