Machine Learning Enabled Wargaming For Theater Undersea Warfare

Our solution leverages a proven, cyber-accredited, Joint, and NATO-deployed wargaming and exercise system already in use at the Naval Warfare Development Command (NWDC) as their wargame adjudication suite.  Our partner Valkyrie’s “Joint Theater Level Simulation – Global Operations” (JTLS-GO) is also used by U.S. INDO-PACOM as their Tier 1 Computer Aided Exercise suite and is employed by the Pacific Warfighting Center (PWC) to develop and validate theater CONOPS.  This multi-domain toolset already includes the necessary asset classes to fully support USW wargaming including satellite sensing and communications, surface and subsurface vessels, vehicles and sensors, weapons and effectors, sonobuoys and strategic sensing, relevant fixed and rotary-wing aviation platforms and systems, countermeasures, weather, and myriad other elements. Skyward has architected a Machine Learning (ML) scenario encapsulation method that enables rapid, consistent, and comparable predictions of scenario outcomes for various strategic and tactical decisions as well as the impact of new technology. The encapsulated scenario can incorporate arbitrary M&S tools at multiple levels of the M&S hierarchy. In this case it would be applied to the Theater Undersea Warfare scenario through the JTLS-GO simulation tool.  The encapsulation also allows for a high-level user interface to assess military worth that can be used independent of the M&S tools needed to create the encapsulated scenario. This approach will empower single operators to rapidly assess a broad expanse of action with appropriate Red Force consideration.  We are also evaluating reinforcement learning (RL) ML techniques to include.  RL agents in the role of wargaming referee or red team player can significantly enhance the wargaming value. Our team will modify the JTLS-GO tool to provide for one or two-player simulations both within Moodle and shipboard environments as requested by the Navy.  We will also incorporate machine learning (ML) approaches, expected to include both Bayesian Neural Nets (BNN’s) for accelerated modeling and Reinforcement Learning (RL) techniques, that will enable our approach by creating Referee and Red Team ML agents.  It will also enable players to gain maximum insight into the impacts of their decisions in the game since a full design space can be quickly run through a BNN. ML explainability will provide coaching to the player on why the decisions produced the results they did and provide guidance on how to potentially improve their decision making.