SBIR: Improving Battle Planning through AI

OUSD (R&E) critical technology area(s): Advanced Computing and Software, Human-Machine Interfaces, Integrated Sensing and Cyber, Trusted AI and Autonomy

Objective: To develop innovative technologies for federated course of action (COA) planning and accelerated COA adjudication using customized reduced order models (ROMs). These ROMs will enable efficient estimation of the modes and eigenvalues of the composition operator, minimizing a physics-informed, objective-based, loss functions for COA evaluation and supporting rapid decision-making in complex battlespace scenarios as well as advanced war-gaming capabilities.

Description: This SBIR topic seeks proposals for developing novel technologies that decouple course of action (COA) adjudication from COA planning through the use of reduced-order models (ROMs). The goal is to construct customized ROMs that generate principal components from simulated and real data sources for estimating the modes and eigenvalues of the composition operator defining the evolution the common operating picture in response to an executed COA. This will enable the development of surrogate models for simulation and war-gaming environments, significantly accelerating the adjudication process. This approach will allow for rapid assessment of numerous COAs generated by potentially disparate planning systems, and advanced war-gaming capabilities for concept development and evaluation.

The proposed technologies should be composable with other similar system models, focusing on the physics of platform movers and effectors, as well as objectives within the battlespace. This composability is crucial for enabling federated planning, where multiple models representing different aspects of the battlespace can be integrated to provide a comprehensive and adaptable planning capability.

A central challenge of this SBIR topic lies in the development of a robust and adaptable methodology for constructing customized ROMs. These ROMs must be tailored to the specific characteristics of diverse battlespace scenarios and objectives. Proposals should detail the specific techniques employed for ROM generation, including the selection of basis functions, discretization methods, and model order reduction algorithms. The proposed methodology should address the challenge of incorporating heterogeneous data sources and varying levels of model fidelity into the ROM construction process. Crucially, the construction process should be demonstrably adaptable, allowing for rapid generation of ROMs specific to new scenarios, platforms, and effectors without requiring extensive retraining or recalibration. The ultimate goal is a ROM construction pipeline that can efficiently produce physics-informed ROMs capable of supporting COA adjudication five orders of magnitude faster than real-time, enabling near-instantaneous evaluation of potential courses of action in dynamic operational environments. This rapid ROM construction capability is essential for maintaining responsiveness and adaptability in the face of evolving threats and objectives.

A critical requirement for the proposed ROM-based adjudication technology is its ability to seamlessly integrate within a federated planning architecture. Proposed solutions must demonstrate how the developed ROMs can be composed with other models, representing diverse aspects of the battlespace, such as enemy behavior, environmental effects, and friendly force capabilities. This composability should enable federated planning across disparate systems and data sources, fostering collaborative decision-making in complex operational environments. A key challenge in federated planning is ensuring data fusion and consistency across different models. Proposals should address how the ROM-based system will handle inconsistencies in data representation, resolution, and timeliness. This includes outlining mechanisms for data validation, conflict resolution, and maintaining a shared understanding of the battlespace across federated models. Furthermore, proposals should describe the interfaces and communication protocols that will facilitate interoperability between the ROM-based adjudication system and other planning components. The end goal is a demonstrably composable ROM technology that seamlessly integrates within a federated planning framework, enabling robust and adaptable COA adjudication across diverse models and data sources, contributing to a more comprehensive and effective planning process. Responsive proposals will address the full DIMEFIL (Diplomatic, Informational, Military, Economic, Financial, Intelligence, and Law Enforcement) spectrum of potential federated ROMs.

Phase I

This SBIR topic is open to Direct-to-Phase II proposals only. Offerors must demonstrate existing technical maturity and feasibility of their approach through preliminary results, prototypes, or prior work. Proposals should clearly articulate the innovation and potential impact of the proposed technology for accelerated COA adjudication and federated planning. A clear transition path to Phase II should be outlined, including existing modeling and simulation environments that would support the rapid development of ROMs for COA adjudication.

Phase II

Phase II efforts will focus on developing and demonstrating a functional prototype of the proposed technology. Offerors will be expected to demonstrate the performance of their ROM-based COA adjudication system using realistic scenarios and data sets. Key performance indicators will include adjudication speed, accuracy of COA evaluation, and scalability in complex, multi-domain environments. Demonstration of composability and integration with existing planning systems will be a critical component of Phase II success. The specific scenarios and evaluation metrics will be finalized in consultation with the Program Manager.

Phase II Base

• Kickoff: Systems requirements review, implementation plan, including identified risks and risk mitigation.
• Month 3: Report on initial prototype, including end-to-end steel thread demonstration and proposed plan for continued implementation. Proposed inventory for initial models and full architectural details including proposed API, and test and evaluation plan.
• Month 6: Documented completion of finalized architecture and planned model development; Detailed continued testing and evaluation plan.
• Month 9: Mid-term performance metrics and demonstration of initial scenario including full federation of planning and adjudication and DIMEFIL effects.
• Month 12: Proposal for extension to full federated system, additional scenario development and current planning and modeling challenges, risks, and mitigation.
• Month 15: Testing and Evaluation Report detailing outcomes in both planning and adjudication including multiple scenarios, DIMEFIL, and multiple domains.
• Month 18: Demonstration of full federated system over multiple scenarios meeting program metrics; Final Report for Phase II to include documentation of current models, limitations, and federated architecture.

All prototypes shall be provided with adequate instructions to support government testing and evaluation using standard equipment and containerized architecture. The components included are expected to meet the program metrics.

Phase III dual use applications

Phase III work is typically oriented towards commercialization of SBIR/STTR research or technology with funding obtained from either the private sector, a non-SBIR/STTR Government source, or both, to develop the technology into a viable product for sale in military or private sector markets. It is envisioned that the technology developed under the SBIR program will have dual-use commercial and DoD applications. In the commercial space, AI-driven model reduction techniques can be applied to improve the performance of simulations, accelerate decision-making, and optimize resource allocation for complex systems. Federated planning and adjudication with reduced order models can be used in healthcare for distributed data analysis, enabling hospitals or research centers to share patient data securely across institutions for more accurate diagnoses or research outcomes. The core technology developed could also allow real-time data processing at the source, optimizing logistics, traffic management, and energy consumption while reducing the reliance on cloud infrastructure.

References

DARPA Broad Agency Announcement, Strategic Chaos Engine for Planning, Tactics, Experimentation and Resiliency (SCEPTER), STO, HR001122S0013

Keywords

Battlefield planning, Course of Action planning, Data Ingest, course of action adjudication, advanced war-gaming, reduced-order models

TPOC-1

DARPA BAA Help Desk

Email

SBIR_BAA@darpa.mil