OSW26BZ02-DV004 — Game-Theoretic AI for Robust Course of Action (COA) Generation and Wargaming
Award Maximum: $2,000,000 Period of Performance: 12 months Phase Type: Direct-to-Phase-II (D2P2)
OBJECTIVE: To develop and demonstrate a mature, scalable, and robust game-theoretic Artificial Intelligence (AI) engine capable of generating and executing novel, optimized courses of action (COAs) in complex, multi-domain, imperfect-information wargaming environments. The objective is to field a capability that consistently outperforms expert human planners and provides decision-makers with a significant strategic advantage in planning, doctrine development, and operational analysis.
DESCRIPTION: Modern military operations are characterized by an astronomically large strategy space, where adversaries' actions are interdependent. Current planning processes are human-intensive, slow, and explore only a "vanishingly small fraction" of possible COAs for both Blue and Red forces. This creates significant operational risk and leaves unexploited opportunities on the table. Standard machine learning approaches are often insufficient as they require massive, labeled datasets that do not exist for future conflicts and frequently produce "black box" solutions that are difficult for commanders to trust, interpret, or certify.
This topic seeks solutions founded in computational game theory capable of computing approximate Nash equilibria in large-scale, zero-sum, imperfect-information games. The desired AI engine will use self-play within high-fidelity simulation environments to learn and refine strategies for both Blue and Red sides simultaneously, without requiring a priori assumptions about adversary tactics.
The proposed solution must demonstrate the following critical attributes:
Dominant Performance: The system must generate COAs that are demonstrably superior to those developed by expert human planners in complex military scenarios. The ability to defeat experienced red teams is the paramount evaluation criterion.
Human-Interpretability: Generated strategies must be transparent and understandable, composed of modular, doctrinally-relevant planning components (i.e., not a monolithic neural network). Commanders must be able to understand the "why" behind the AI's recommendations.
Scalability: The AI architecture must be capable of scaling from tactical engagements (e.g., individual flight combat) to operational-level scenarios involving thousands of assets across multiple domains (air, sea, land) and extended time horizons.
Computational Efficiency: The solution should operate effectively on modest computational footprints (e.g., single or small-cluster CPU-based workstations), avoiding reliance on cost-prohibitive, large-scale GPU clusters for its core training and inference loops.
"Anytime" Capability: The algorithm must be capable of providing a valid, usable strategy at any point during its computation cycle, with the solution quality improving as more time and resources are allocated.
PHASE I: This topic is accepting Direct to Phase II proposals only. Strong proposals should document prior experience:
A detailed white paper describing the underlying computational game-theory model and algorithmic approach used to find and refine strategies in imperfect-information games.
Demonstrated results (e.g., data, reports, or videos) of the AI's performance against expert human teams or other state-of-the-art AI benchmarks in a complex, multi-domain wargaming or simulation environment. Performance must be quantified using a clear utility or scoring function.
Evidence of the AI's ability to generate novel, effective, and human-interpretable strategies (e.g., examples of generated COAs).
Technical specifications detailing the computational resources required for both training and execution, and data supporting claims of scalability from small to large-scale scenarios.
Proposals lacking sufficient evidence of a mature, existing prototype and demonstrated performance will be deemed non-responsive.
PHASE II: Phase II will focus on:
Building on the proven feasibility from Phase I, the offeror will mature, harden, and scale their prototype AI engine for defense-specific applications. The scope of work will include:
Integrating the AI engine with a government-designated modeling and simulation (M&S) environment (e.g., Command: Professional Edition (CPE), AFSIM, or others).
Conducting a series of validation and verification (V&V) events in government-provided scenarios of increasing complexity, including multi-domain swarm scenarios and joint all-domain operations.
Systematically demonstrating the robustness of the AI by assessing its performance under conditions of degraded communications, sensor uncertainty, incomplete information, and novel adversary tactics not present in the training set.
Delivering a robust, containerized software prototype of the AI engine and a technical data package (TDP) sufficient for government use in wargaming, analysis, and COA development.
PHASE III DUAL USE APPLICATIONS:
Government/Military: The primary application is to serve as a core component of next-generation wargaming centers, operational planning cells, and training programs across the DoD and the Intelligence Community. It can function as a "blue" planning aide, an ultra-capable "red" opponent for training, or an impartial adjudicator for COA analysis.
Commercial: The underlying game-theoretic reasoning engine can be adapted for a wide range of commercial markets that involve high-stakes strategic interaction under uncertainty. These include financial market modeling, cybersecurity defense strategies, complex business negotiations, supply chain optimization, and resource allocation in competitive environments. The offeror is expected to pursue these commercial applications to ensure long-term viability and innovation.