The growing interest in situationally aware, tactically intelligent computer-generated forces (CGF) for military training underscores the necessity for a conceptual framework using interpretable AI methods to develop such behavior. By prioritizing system interpretability, an effective debriefing of training exercises is made possible.

Our concept to generate such tactically intelligent behavior involves three essential processes: Situation Analysis, providing information of the current situation; Risk-/Opportunity-Assessment, evaluating the information; and Decision-Making and Planning, utilizing the assessment for tactical decisions. A critical challenge in simulating these processes is formulating comprehensible risks and opportunities in team-based situations.

In this paper, we present details on the Risk-/Opportunity-Assessment process, to be applied on a beyond-visual-range (BVR) air combat scenario within an agent-based simulation. The proposed solution is comprised of three key steps. First, qualitatively and semi-quantitatively assigning risks and opportunities to a situation based on the relations between all entities. Second, a full quantification of the identified risks and opportunities, considering team and entity perspectives. Finally, juxtaposing and evaluating the results to formulate an interpretable outcome.

After describing our conceptual framework, we introduce a methodology for evaluating the conditions of one's own entities. This evaluation focuses on assessing an entity’s readiness and capacity to engage in combat operations, independent of the opposing forces. The primary challenge resides in generating a reproducible and interpretable outcome, which incorporates various factors including individual team member proficiency, equipment status, and overall team cohesion.

The effectiveness of the proposed methodology was evaluated by implementing it within an experimental framework that simulated CGF in a BVR training simulator. The resulting outcomes were compared and benchmarked against tasks executed by subject matter experts and will be included in the paper.

Luca Winkler

Affiliation

Refine Results

Computer-Generated Forces Team Behavior within Air Combat Simulations: Concept and Agent Structure