Computer Generated Forces (CGF) enable training sessions to be conducted in computer environments which may involve only a few individuals who operate as if they were training in environments which include a large number of individuals and manned or robotic military units. A recent National Academy of Sciences report defines human behavior representation as a computer based model that mimics the behavior of a single human or the collective behavior of a team of humans. The same report decries the lack of behavior realism in military situations. This paper addresses a key question of how to model units or simulated individuals, or manned vehicles, when these entities have the ability to learn from experience. The behavior of CGF cannot be considered to be purely static, but will necessarily evolve with experience, just as the "human student" being trained in this artificial environment also learns through experience. We discuss a mathematically based framework for Learning Behavior Representation based on stochastic automata, which are randomized finite or infinite state machines, controlled by random neural networks with learning algorithms. This approach can lead to more realistic capabilities for computer based simulation and training systems.
展开▼