Design, Reconfigure, and Evaluate Autonomous Models in Training

ABSTRACT: The proliferation of autonomous- and human-machine systems necessitates the creation of new tools for system design and evaluation. Key among these are simulation testbeds that support interactions between multiple warfighters and systems, and methods for creating and integrating intelligent agent models into simulation environments. We see a significant opportunity to advance the state of the art in human-machine systems, including the efficiency of developing intelligent agent models, the integration of agent representations into simulation environments, and the speed of evaluating human-machine systems using simulation environments. We refer to our solution as DREAMIT Design, Reconfigure and Evaluate Autonomous Models In Training. DREAMIT centers on two concepts. The first is to combine multiple intelligent agent models, specified at different levels of abstraction, within the simulation environment. By leveraging different levels of abstraction, agent models will produce behavior at the desired levels of fidelity without incurring undue computational complexity and cost. The second concept is to create a virtual data loop. The aim is to capture data from training simulations, to apply computational methods to the data in order to facilitate agent development, and to close the loop by gathering additional data after re-embedding agent models in the simulation environment.; BENEFIT: In creating DREAMIT, we expect to deliver a best-in-class solution to satisfy the Air Forces needs to conduct multi-agent simulations, to accelerate the pace of developing intelligent agent models, and to evaluate human-machine systems more efficiently. The major benefits of DREAMIT are: (1) reconfigurability, in terms of the exercise scenario, and the constituent live and simulated players; (2) adaptability of intelligent agent models based on training history and encountered tactics; (3) anticipation of Red and Blue Force responses to new tactics based on intelligent agent models; and (4) efficient and economical evaluation of alternate human-machine system designs. These features will make DREAMIT applicable in other domains, military and commercial alike, where flexible scenario construction and adversary prediction are needed for training; for example, in other military services, for emergency responders, and in security response coordination.