DoD plans to acquire new simulators for training warfighters to operate in hostile urban environments. These simulators need computer-controlled urban-oriented entities to interact with trainees realistically, taking into account: physical models of human movement and performance degradation; reactive behaviors for selecting immediate actions; and reasoning for providing situation assessment and planning. To foster re-use of these entities, their software models must be stored in a repository. Traditional CGFs' modeling approaches, oriented towards representing platforms, not people, are inappropriate to meet these challenges. We propose a divide-and-conquer approach. We separate the repository and the UHE simulation from the training simulator to foster re-use. We further divide the UHE representations into behaviors and bodies, each of which is stored in the repository. We use existing COTS products to implement the body, and use the ONR-funded Subsumption Architecture developed at MIT's AI Lab to implement the behaviors. The resulting repository will contain UHE behaviors that are more realistic, more efficient, more numerous, and easier for subject matter experts to audit, than competing approaches. Reliance on COTS software lowers cost, time, and risk. We propose to obtain sample individual and group UHE behaviors from the sponsor. We will classify these sample behaviors according to Subsumption Architecture levels and test them in its Behavior Language environment. We will define sensor and actuator APIs so that other CGFs can re-use the behaviors written to that API. At the customer's option, we will retarget the back end of the BL compiler to be compatible with C/C++ code. Finally, we will document our findings in a Final Report.
