Virtual Battlespace Behavior Generation through Class Imitation

Military organizations need realistic training scenarios to ensure mission readiness. Developing the skills required to differentiate combatants from non-combatants is very important for ensuring the international law of armed conflict is upheld. In Simulated Training Environments, one of the open challenges is to correctly simulate the appearance and behavior of combatant and non-combatant agents in a realistic manner. This thesis outlines the construction of a data driven agent that is capable of imitating the behaviors of the Virtual BattleSpace 2 behavior classes while our agent is configured to advance to a geographically specific goal. The approach and the resulting agent promotes and motivates the idea that Opponent and Non-Combatant behaviors inside of simulated environments can be improved through the use of behavioral imitation.