Framework for Intelligent Simulation Command with Hierarchically Embedded Reinforcement Learning (FISCHER) Navy SBIR FY2018.1 Sol No.: Navy SBIR FY2018.1 Topic No.: N181-083 Topic Title: Framework for Intelligent Simulation Command with Hierarchically Embedded Reinforcement Learning (FISCHER) Proposal No.: N181-083-0497 Firm: DECISIVE ANALYTICS Corporation 1400 Crystal Drive Suite 1400 Arlington, Virginia 22202 Contact: Jonathan Day Phone: (703) 414-5015 Web Site: http://www.dac.us Abstract: Wargaming plays a critical role in the modern military, both as a decision support tool for command and control centers and as a training tool for developing the future force. In both of these applications it is valuable to have access to highly skilled automated actors. In a decision support context, this facilitates fast, high fidelity simulations with a variety of possible battlefield conditions. In a training context, this supports interactive learning tools and allows commanders to hone their skills through competition with difficult opponents. Historically, Artificial Intelligence agents that can match or beat expert human performance have been limited to turn-based games. However, due to the success of IBM�?Ts Deep Blue system at Chess, and more recently, DeepMind�?Ts AlphaGo system at Go, the latest research has begun to investigate if the techniques employed by these systems can be improved to achieve expert-level performance at less well-defined, real-time games. To manage the fundamental complexities that military-domain AI systems face, DECISIVE ANALYTICS Corporation proposes the Framework for Intelligent Simulation Command with Hierarchically Embedded Reinforcement Learning (FISCHER). FISCHER will leverage the latest developments in Game Theory and Reinforcement Learning to provide a wargaming artificial intelligence that can perform at a humanlike level. Benefits: The primary goal of this research effort is the development of the Framework for Intelligent Simulation Command with Hierarchically Embedded Reinforcement Learning (FISCHER) prototype. FISCHER will provide a human-level decision making Artificial Intelligence for wargaming applications. This will improve command and control decision support systems by improving the fidelity of their underlying wargaming simulations, enabling more realistic evaluations of potential mission outcomes. FISCHER will also support the development of interactive command training tools that will help train the next generation of warfighters. The technologies developed under this SBIR are also highly applicable to commercial products relying on agent-based simulation and to the commercial gaming industry. Commercial game developers and e-sport organizations could benefit from intelligent AI systems to help optimize game balance or test novel strategies. Return