Configurable Adversary Response Prediction: Building Efficient Expectation Models from High-Fidelity Behavior Simulations

摘要

The Air Force has an interest in a run-time mission Pilot's Assistant (PA) that will support tactical pilots in the rapid assessment of information quality and reconsideration of decisions that the information supports. Such a capability requires an efficient, predictive knowledge base that enables rapid situation assessment and decision support. The Configurable Adversary Response Prediction (CARP) project addresses two key technical challenges to the development of the PA. The first is to extend and exploit the state of the art in modeling and simulation, particularly in the modeling of human decision making, to support simulation of scenario and mission outcomes that provide the analytical forecasts necessary to perform situation assessment. The second is to represent the results of these analyses in an efficient knowledge base that can create assessments in real time, overcoming the difficulty of running large-scale analyses during mission execution. The output of the CARP application will be sets of Behavior Envelopes that align expected observations with assumptions about adversary goals and tactics. These envelopes will provide a knowledge base that allows the PA to rapidly identify mismatches between assumptions, awareness, and observations. CARP will produce this efficient knowledge base through collaborative exploration of predictive behavior spaces, using a guided user interface for experimenting with realistic simulations and summarizing their predictive outcomes. CARP's simulation testbed will be a robust integration and adaptation of predictive simulation systems and models. The testbed used by CARP will exploit the most accurate available models, together with tools to support and adapt model parameters to improve predictions, and then to aggregate simulation results into Behavior Envelopes, which encapsulate expected observed behaviors in varied situations. To prove the concept, we are working in collaboration with TiER I and EduWorks, who are developing innovative technologies for high fidelity simulation-based modeling of adversary decision making. The paper describes our work to date, including challenges met and advances made. It also discusses our integration plans for the tactical environment evaluations, which will be the capstone event to the efforts.