SUMMARY

Operating in contested environments against near-peer and/or peer adversaries has the potential to place overwhelming demands on operational planners and warfighters, thereby requiring decision support assistance. Recent advancements in the use of Artificial Intelligence (AI) as video game-playing agents may have applications within the battlespace management domain.

Patriot Labs is seeking innovative solutions for applying State-of-the-Art Artificial Intelligence (AI) and Machine Learning (ML) approaches to battle management. Game playing technologies have the potential to become both a valuable planning tool, useful for performing “what-if” analysis of plans and courses of action, as well as a means of providing fast response capability for execution management.

For purposes of this CFI, proposed capabilities should enable real-time reasoning based on battlespace developments. Approaches should be capable of assisting planners and decision-makers with identifying and executing optimal courses of action during wartime engagements. Key capabilities should include: (i) AI algorithms and ML methods applicable to deriving strategies, supporting complex operations, and generating responses in adversarial environments; (ii) functionality for capturing human expertise, learning from subject matter experts, and augmenting warfighter capability through interactive gameplay based on operational use cases; and (iii) the application of domain adaptation techniques to transfer AI strategies and/or machine-learned models into relevant challenge/problem simulations.

Potential approaches may include the development of AI approaches in unclassified game-play domains and methods for transferring derived AI capabilities to either unclassified or classified scenarios. Special consideration given to proposals that leverage existing game-playing engines to newly developed interactive “simulation” environments based on user-specified battlespace scenarios. Anticipated technical challenges that will need to be addressed in the solution include representation of the game environment, dealing with imperfect information, developing long-term strategy, learning from limited data, learning via interaction with human planners, and domain adaptation and transfer.