Data-driven behavior modeling for computer generated forces – a literature survey

Computer Generated Forces (CGFs) have been used within military simulation-based, training and decision support applications for decades. CGFs are autonomous or semi-autonomous entities that typically represent military units such as tanks, soldiers, and combat aircrafts. Their main purpose, at least within training applications, is to reduce human role playing, which allows for more efficient use of military training facilities. Although CGFs are undoubtedly useful, their behavioral capabilities are often limited to follow doctrines, rules of engagement or heuristics identified by human domain experts that not necessarily represent the behavior observed in the CGF’s real-world counterpart. In this report we introduce and provide a literature review of works related to the Data-Driven Behavior Modeling (DDBM) approach, which is an alternative to the traditional domain expert based behavior modeling approach. In DDBM computers are employed to analyze and identify behavioral patterns in data using machine learning techniques. DDBM is believed to more efficiently produce CGFs that are more objective and realistic compared to CGFs where the behavioral patterns mainly have been identified using subjective human experts. This report is the result of a collaborative effort between the Norwegian Defence Research Establishment (FFI) and the Swedish Defence Research Agency (FOI), "Technical arrangement number 4 FFI-FOI - Terrain Analysis and Synthetic Agents".