Representational Approach to Knowledge and Multiple Skill Levels for BroadClasses of Computer Generated Forces

摘要

Current computer generated forces (CGFs) in the 'synthetic battlespace', atraining arena used by the military, exhibit several deficiencies. Human actors within the battlespace rapidly identify these CGFs and defeat them using unrealistic and potentially fatal tactics, reducing the overall effectiveness of this training arena. Simulators attached to the synthetic battlespace host local threat systems, leading to training inconsistencies when different simulators display the same threat at different levels of fidelity. Finally, current CGFs are engineered 'from the ground up', often without exploiting commonalities with other existing CGFs, increasing development (and ultimately training) costs. This thesis addresses these issues by proposing a domain-independent design methodology and a supporting software architecture for the Distributed Mission Training Integrated Threat Environment (DMTITE). This architecture uses approaches from software engineering and database management and identifies an extensible knowledge representation to support CGFs in various domains (land, surface, and air), shifting development efforts from 'structure implementation' to 'knowledge implementation' CGFs developed using this paradigm also have access to domain-independent features such as skills vectors and a combat psychology model, which act as a time-limited Turing test by making CGF behaviors unpredictable (but not random) and believable.