DSTO Doctrine to Code Project

The Doctrine to Code Project for DST Group (formerly DSTO) is investigating novel methods for supporting the full process of executing combat simulations, starting from the textual documentation describing the requisite behaviours. It involves the development of a general behaviour meta-model for representing the scenario descriptions, the use of natural language processing to extract behaviour models from text, and model transformations to generate documentation and code from the models.

Researchers: Dr. Matt Selway, Prof. Markus Stumptner, Dr. Georg Grossmann, Dr. Wolfgang Mayer


The Doctrine to Code Project aims to reduce the time taken for DST Group to develop behavioural models for combat simulations and to improve the reuse of such behavioural models across simulation environments. To achieve this, the project has three main components:

  1. a general behaviour meta-model and notation that can represent reactive entity behaviour independently from any particular simulation environment;
  2. natural language processing to generate executable behaviour models from doctrinal texts
  3. model transformations for the generation of executable code and documentation

These three components are outlined below.

Behaviour Meta-Model

There is a need to reduce the time taken to develop entity behaviours for simulation scenarios, improve the consistency with which the models are developed, and increase the reuse of these behaviours across different simulation environments. To that end we have developed a generic behaviour meta-model and graphical notation (called the Hierarchical Behaviour Model and Notation, or HBMN for short) that allows users to define all the required behavioural elements, e.g., sequential, parallel, and decision behaviours, as well as data flows, resources, etc. In addition, the behaviour meta-model incorporates hierarchical decomposition and reusability mechanisms to allow the creation of more complex behavioural models from simpler models in a top-down or bottom-up fashion. Moreover, behaviour models can be stored in a library for reuse in other models that are developed later.

Process Extraction from Text

To assist in the development of behaviours based on doctrine, this project involves the identification and extraction of behaviour specifications and processes from text. By automating this knowledge elicitation process, the time taken to develop doctrine based behavioural models will be reduced and the understanding and communication of the behaviours will be improved. Moreover, by taking a novel knowledge-based approach to the natural language processing, we are able to improve the extracted models over time as reusable behaviours are developed. That is, the text analysis can identify instances of existing behaviours in the text and incorporate them into newly extracted models. Therefore, as the library of behaviour models grows, the usefulness and correctness of extracted models is improved.

Model Transformation to Code

Further reducing the time taken to model, prepare, and run simulations requires model transformation and code generation capabilities. This is addressed by the third component, whereby we are developing methods of refining and mapping the generic behaviour models into simulation environment specific models and executable code. This involves the semi-automated selection and composition of appropriate behaviour definitions into complete valid models, followed by their mapping and transformation into specific programming languages and the APIs used by a specific simulation environment. Moreover, additional transformations automate the generation of documentation about the behavioural models, for reporting and communication purposes.




  • [PDF] [DOI] M. Selway, K. R. Owen, R. M. Dexter, G. Grossmann, W. Mayer, and M. Stumptner, “Automated techniques for generating behavioural models for constructive combat simulations,” in Data and decision sciences in action: proceedings of the australian society for operations research conference 2016, 2018, p. 103–115.
    author = {Matt Selway and Kerryn R. Owen and Richard M. Dexter and Georg Grossmann and
    Wolfgang Mayer and Markus Stumptner},
    title = {Automated Techniques for Generating Behavioural Models for Constructive Combat
    booktitle = {Data and Decision Sciences in Action: Proceedings of the Australian Society for Operations Research Conference 2016},
    pages = {103--115},
    year = 2018,
    publisher = {Springer},
    series = {LNMIE},
    doi = {10.1007/978-3-319-55914-8_8},