Automated tactics modeling: techniques and applications

Interactive computer simulation can be an efficient and effective tool for training, but faces several technical obstacles: first, agents in the scenario (simulated allies and opponents) must exhibit realistic tactics; second, the dynamics of the simulation must be valid in order to reinforce tactics which are applicable in reality; and third, the training system should provide corrective feedback when students commit errors. To be practical, a technical approach must accomplish all of these objectives quickly and inexpensively. In this dissertation I develop solutions for these requirements with techniques that enable subject matter experts (SMEs) to create models of their own tactics. Each of the three obstacles is addressed with an approach based on behavior modeling: first, the models provide a way to transfer real-world tactics to software agents. Second, the dynamics of the simulator are validated by measuring the effectiveness of real-world vs. synthetic tactics in the simulator. Third, students using the simulator for training are evaluated by comparing their actions to those predicted by models of SME behavior. Because the models are constructed by an automated process, knowledge engineers and software developers need not mediate between the SME and the student audience. The model itself becomes a medium for indirect educational interactions between the SME and student.