This thesis develops and systematically investigates a modular, partly probabilistic Situation Assessment Process (SAP) for computer-generated Red Force agents in BVR air combat training simulations. The objective of this work is to provide a reliable, explainable, and evaluable interpretation of the tactical intentions of adversarial (Blue Force) entities in order to enable CGF agents to exhibit situation-adaptive and tactically intelligent behavior. In this way, Blue Force trainees can be realistically challenged during simulation without manual intervention by an instructor. Based on domain-specific requirements, a multi-stage concept is presented that integrates object grouping, distributed evidence and hypothesis fusion, uncertainty representation, and tactic recognition. The process is implemented within a distributed multi-agent architecture using tactical templates that can be edited by Subject-Matter Experts (SMEs), confidence-based weighting, and a traceable history of the internal SA decision-making process across multiple cooperating Red Force agents. The experimental validation in constructive and virtual simulations shows that the developed SAP (i) robustly groups tactically interacting entities, (ii) consistently fuses maneuver evidence across variations in distance, fusion method, and agent-specific confidence factors, (iii) infers tactical intentions under uncertainty, with the resulting uncertainty structure being closely coupled to the stability of maneuver fusion, and (iv) generates reproducible decisions about adversarial intentions under uncertainty. As part of the validation, these decisions are compared with the situation assessments of human operators and yield comparable results. Thus, this dissertation contributes to research gaps concerning integrated and explainable SA processes for CGF-Red Force agents, where existing approaches have so far combined uncertainty representation, competing tactical hypotheses, and distributed multi-agent information fusion only to a limited extent within a coherent process. It further addresses the need for modular, SME-adaptable SA architectures with traceable internal state representation and for reproducible evaluation methods to assess SA correctness, confidence calibration, and temporal development in BVR training simulations. The SAP is integrated into a modular CGF-Red Force simulation framework and is used for agent control as well as training assessment. The identified limitations indicate further research needs, particularly with regard to dynamic SA parameterization, real-time-capable inference chains, and adaptive MEBN models.     «

This thesis develops and systematically investigates a modular, partly probabilistic Situation Assessment Process (SAP) for computer-generated Red Force agents in BVR air combat training simulations. The objective of this work is to provide a reliable, explainable, and evaluable interpretation of the tactical intentions of adversarial (Blue Force) entities in order to enable CGF agents to exhibit situation-adaptive and tactically intelligent behavior. In this way, Blue Force trainees can be real...     »