Virtual Agent for Data Fusion and Understanding

Navy SBIR 25.2 - Topic N252-089
Naval Air Systems Command (NAVAIR)
Pre-release 4/2/25   Opens to accept proposals 4/23/25   Closes 5/21/25 12:00pm ET
   [ View Q&A ]

N252-089 TITLE: Virtual Agent for Data Fusion and Understanding

OUSD (R&E) CRITICAL TECHNOLOGY AREA(S): Trusted AI and Autonomy

OBJECTIVE: Conceive and develop a novel artificial intelligence (AI) system—virtual agent—capable of sophisticated data fusion and comprehension, adapted for the Navy’s diverse data ecosystem. This virtual agent will be competent to process, explain, and generate actionable intelligence from heterogeneous data sources; thereby augmenting situational awareness and decision-making acumen within the dynamic maritime theater.

DESCRIPTION: The Navy is inundated with data emanating from myriad multiform sources: sensor data from maritime and aerial platforms, intelligence dossiers, maintenance logs, environmental metrics, communications intercepts, and so forth. Human analysts are presented with complex radar and sonar returns mapping physical spaces and threats, detailed textual and visual intelligence reports necessitating advanced linguistic and visual analytics, operational and maintenance data indicative of asset readiness, and critical oceanographic and meteorological data conditioning strategic operations.

The colossal volume and diversity of these data pose considerable challenges in terms of real-time processing, comprehensive understanding, and the distillation of actionable intelligence. Prevailing approaches anchored in generative AI have exhibited only limited success in natural language and image production and are ultimately wrecked on the shoals of complexity that human-level/human-style intelligence uniquely can comprehend. Generative AI models are constitutionally defective by their inability to explain—and not merely predict—patterns in data, which in addition disables them from generalizing across disparate data types.

To transcend these limitations, a paradigm shift towards a hybrid AI approach—synergizing human-style machine learning (ML) with human-style symbolic AI into a neurosymbolic hybrid—is imperative. Symbolic AI, using knowledge graphs, ontologies, and rule-based systems, can endow the virtual agent with domain knowledge and reasoning faculties. This consilience of ML and symbolic AI will empower the agent to integrate and interpret a wide spectrum of data, discern explicit causality from latent correlations, and generate robust, actionable insights. The envisioned virtual agent must thus leverage both advanced non-Large Language Models (non-LLM) ML and symbolic AI to actualize comprehensive data fusion and deep understanding.

Work produced in Phase II may become classified. Note: The prospective contractor(s) must be U.S. owned and operated with no foreign influence as defined by 32 U.S.C. § 2004.20 et seq., National Industrial Security Program Executive Agent and Operating Manual, unless acceptable mitigating procedures can and have been implemented and approved by the Defense Counterintelligence and Security Agency (DCSA) formerly Defense Security Service (DSS). The selected contractor must be able to acquire and maintain a secret level facility and Personnel Security Clearances. This will allow contractor personnel to perform on advanced phases of this project as set forth by DCSA and NAVAIR in order to gain access to classified information pertaining to the national defense of the United States and its allies; this will be an inherent requirement. The selected company will be required to safeguard classified material during the advanced phases of this contract IAW the National Industrial Security Program Operating Manual (NISPOM), which can be found at Title 32, Part 2004.20 of the Code of Federal Regulations.

PHASE I: Conceive and develop a neurosymbolic hybrid virtual agent. This initial phase comprises the architectural design of the system and the selection of data types (including, but not limited to radar and sonobuoy data). The objective is to construct a nascent model that validates the feasibility of integrating these diverse data streams and generating preliminary insights. Additionally, this phase will encompass the establishment of evaluative metrics of system performance and the formulation of a scalable development plan for subsequent phases. Reference sources to determine trust/confidence in the system will include external validated knowledgebases and domain-specific training.

The Phase I effort will include prototype plans to be developed under Phase II.

PHASE II: Build upon the foundational insights of Phase I, advance toward the comprehensive development, rigorous testing, and empirical validation of the virtual agent. Build and refine the algorithms and models, integrate additional data sources, and enhance the prototype system’s real-time processing capabilities. Test the virtual agent within realistic Navy operational scenarios to assess its efficacy in terms of accuracy, alacrity, and resilience. Demonstrate the agent’s prowess in delivering holistic situational awareness, anticipating potential threats, and proffering actionable strategic recommendations. Ensure that the virtual agent is a deployable system that substantially augments the Navy’s data fusion and understanding capabilities, thereby elevating operational efficacy and strategic decision-making.

Work in Phase II may become classified. Please see note in Description paragraph.

PHASE III DUAL USE APPLICATIONS: Complete final testing. Perform necessary integration and transition for use in operational applications with appropriate platforms and agencies, and future combat systems under development.

Commercially, this product could be used to enable security monitoring, smart city operations center, power grid monitoring, and wherever large amounts of sensors or inputs are utilized.

REFERENCES:

  1. Garcez, A. D. A.; Gori, M.; Lamb, L. C.; Serafini, L.; Spranger, M. and Tran, S. N. "Neural-symbolic computing: An effective methodology for principled integration of machine learning and reasoning." Journal of Applied Logics — IfCoLog Journal of Logics and their Applications, 2019. https://arxiv.org/pdf/1905.06088
  2. Hoy, M. B. "Alexa, Siri, Cortana, and More: An Introduction to Voice Assistants." Medical Reference Services Quarterly, 37(1), 2018, pp.81-88. https://pubmed.ncbi.nlm.nih.gov/29327988/
  3. Marcus, G. and Davis, E. "Rebooting AI: Building artificial intelligence we can trust.". Pantheon Books, New York, 2019. https://search.worldcat.org/formats-editions/1083223029
  4. "National Industrial Security Program Executive Agent and Operating Manual (NISP), 32 U.S.C. § 2004.20 et seq. 1993". https://www.ecfr.gov/current/title-32/subtitle-B/chapter-XX/part-2004

KEYWORDS: Data fusion; artificial intelligence; machine learning; AI/ML; virtual agent; neurosymbolic agent; decision-making; symbolic AI

** TOPIC NOTICE **

The Navy Topic above is an "unofficial" copy from the Navy Topics in the DoD 25.2 SBIR BAA. Please see the official DoD Topic website at www.dodsbirsttr.mil/submissions/solicitation-documents/active-solicitations for any updates.

The DoD issued its Navy 25.2 SBIR Topics pre-release on April 2, 2025 which opens to receive proposals on April 23, 2025, and closes May 21, 2025 (12:00pm ET).

Direct Contact with Topic Authors: During the pre-release period (April 2, 2025, through April 22, 2025) proposing firms have an opportunity to directly contact the Technical Point of Contact (TPOC) to ask technical questions about the specific BAA topic. The TPOC contact information is listed in each topic description. Once DoD begins accepting proposals on April 23, 2025 no further direct contact between proposers and topic authors is allowed unless the Topic Author is responding to a question submitted during the Pre-release period.

DoD On-line Q&A System: After the pre-release period, until May 7, 2025, at 12:00 PM ET, proposers may submit written questions through the DoD On-line Topic Q&A at https://www.dodsbirsttr.mil/submissions/login/ by logging in and following instructions. In the Topic Q&A system, the questioner and respondent remain anonymous but all questions and answers are posted for general viewing.

DoD Topics Search Tool: Visit the DoD Topic Search Tool at www.dodsbirsttr.mil/topics-app/ to find topics by keyword across all DoD Components participating in this BAA.

Help: If you have general questions about the DoD SBIR program, please contact the DoD SBIR Help Desk via email at [email protected]

Topic Q & A

4/25/25  Q. Is it necessary to include a Phase I Option plan for all proposals, or only proposals that are specifically applying for both a Phase I period of performance and a Phase I Option?
   A. As indicated in the Navy SBIR 25.2 Phase I instruction, the Phase I Option period furthers the effort in preparation for Phase II and bridges the funding gap between the end of Phase I and the start of Phase II. A selected proposal that has included no Option funding or tasks in their submission will remain unfunded during this gap in time. Please review the Navy SBIR 25.2 Phase I instruction Phase I Submission Instructions for further details on requirements.
4/23/25  Q. 1. What level of symbolic reasoning is expected in Phase I—e.g., causal reasoning, rule-based inference, ontology-guided search—and are there specific ontologies or knowledge graphs already used by the Navy that should be leveraged or extended?

2. How should the virtual agent present its reasoning or derived insights to analysts or commanders? Are natural language explanations, decision trees, or visual traceability paths expected for model transparency and trust?

3. What specific evaluative metrics will be used to gauge success during Phases I and II—e.g., fusion latency, insight accuracy, false positive rates, user trust scores, threat detection rates—and will any human-in-the-loop assessments be included?

4. How do you envision human analysts interacting with the virtual agent in decision-making workflows? Will the agent offer recommendations, request clarifications, or function autonomously in any scenario?

5. The solicitation specifies a preference for non-LLM machine learning approaches. Are there specific ML paradigms (e.g., graph neural networks, probabilistic reasoning, hybrid rule learning) that align with current Navy research directions or should be avoided?
   A.
  1. The intent is to have the contractor start out with fresh sheet without bias (conceive and develop a novel artificial intelligence (AI) system).
  2. Natural language explanations, decision trees, or visual traceability paths are methods to provide human style explanation to gain trust in the data fusion abilities and relationships of the virtual agent?
  3. For Phase I will be based on the contractor’s ability to conceive and develop a novel artificial intelligence (AI) system— virtual agent—capable of sophisticated data fusion and comprehension. The architecture, data flow, and approach being taken will form the basis of the evaluation. In Phase II, there is the potential of human-in-the-loop assessment.
  4. This will depend on the how the analysts and virtual agent interact. Individuals can either be auditory, visual or both consumers of information. The comprehension and explanation of the information being presented is part of the overall concept of this novel system approach. Based on the evaluation criteria noted in Q3 should the system be able to minimize the operator cognitive workload while enhancing the overall operational efficiency.
  5. The intent is to have the contractor start out with fresh sheet without bias (conceive and develop a novel artificial intelligence (AI) system). Symbolic AI, using knowledge graphs, ontologies, and rule-based systems, can endow the virtual agent with domain knowledge and reasoning faculties.
4/22/25  Q.
  1. Is it acceptable to use LLMs as a base model if symbolic inference and explainability layers are prioritized?
  2. Are specific data formats or types prioritized (e.g., sonar, radar, environmental logs, SIGINT)?
  3. What constitutes sufficient symbolic reasoning in the Navy’s view: rule-based reasoning, ontologies, causal inference?
  4. Would the Navy support integration with commercial knowledge graphs or require custom-built taxonomies?
    5. ,/ol>
   A.
  1. The distribution and prioritization of the various compositions, ontologies, methods and human explainability is left up to the proposer to determine. It is noted in the topic that “synergizing human-style machine learning (ML) with human-style symbolic AI into a neurosymbolic hybrid—is imperative”
  2. For the Phase I effort, the contractor is expected to utilize whatever information or simulation data to illustrate their proposed solution. This is to provide the contractor with the greatest leeway with information/data sets they have on hand.
  3. Sufficient symbolic reasoning will be based on the robustness of the proposed conceptual solution and envisioned implementation presented. Of note is that the training datasets and databases to train “the system” are limited hence the rationale for the symbolic reasoning preference.
  4. The Navy is interested in a functional hybrid system however the composition is determined by the Contractor.
4/21/25  Q. Sensor data streams vary by multiple orders of magnitude. Will you be able to furnish some rough order of magnitude estimations on how many perspective systems for integration will stream data per second on the order of megabytes, gigabytes, or terabytes…?
   A. Data rates depend on the specific system. Figure in the megabyte to gigabyte range.
4/21/25  Q. Can you list the systems that were deemed insufficient for this purpose?
   A. Not sure what is referred to or meant by "deemed insufficient"
4/21/25  Q. What sample datasets are you prepared to provide?
   A. None. For Phase I the contractor will be using simulated or data they currently have on hand.
4/21/25  Q.
  1. What level of system autonomy or human-in-the-loop decision making is envisioned for the virtual agent in operational environments (e.g., supervisory control, fully autonomous, recommendation-only)?
  2. Is there a preferred or recommended set of data types or domains beyond radar and sonobuoy data (e.g., EO/IR, maintenance logs, SIGINT) that offer the greatest near-term value to the government?
  3. Are there existing Navy ontologies, knowledge graphs, or curated knowledgebases that offer a reference point for Phase I symbolic reasoning development?
  4. What metrics or benchmarks does the government anticipate using to evaluate the success of neurosymbolic reasoning and cross-domain data integration in Phase I?
  5. Will the government provide access to synthetic or representative datasets during Phase I, or is the offeror expected to generate all test data independently?
  6. Should the Phase I architecture account for real-time streaming data inputs, or is batch ingestion and reasoning sufficient for initial feasibility assessment?
  7. Are there any legacy systems, platforms, or C4ISR tools the virtual agent should eventually interoperate with or enhance as part of long-term transition planning?
  8. For purposes of Phase II planning, are there specific classified operational environments or combat systems in which this capability is envisioned to transition?
   A. 1. This will be determined by the effectiveness of the proposed solution. The goal is reduction human cognitive workload while making the options behind the decision(s) and prediction(s) (forecast) provided be described by reasoning that has been applied. Clear and concise explanations which build trust (repetitive similar solutions provided without input variation) to the operator.

2. This virtual agent is anticipated to operate over the entire mission capability of the platform – from maintenance through tactical operations (such as sensor data from maritime and aerial platforms, intelligence dossiers, maintenance logs, environmental metrics, communications intercepts, etc.)

3. The intent is to have the contractor start out with fresh sheet without bias (conceive and develop a novel artificial intelligence (AI) system).

4. Phase I. Is based on the approach, architecture and implementation as described in the Phase I proposal. You will be evaluated on the relationship between the outcomes suggested in the proposal and performed in Phase I.
Phase II. The system should not degrade current performance metrics, increase operator’s cognitive workload, misinform decision-making (hallucination). Processing time should be faster than human performing the same function. They are prioritized equally (speed and accuracy)..

5. For the Phase I effort, the contractor is expected to utilize whatever information or simulation data to illustrate their proposed approach and solution is viable. This is to provide the contractor with the greatest leeway with information/data sets they have on hand.

6. It is envisioned that the virtual agent will eventually doing both. For Phase I, batch ingestion and reasoning sufficient for initial feasibility is acceptable with the understanding that real-time data streaming as a final product shall be described as a Phase I output.

7. This will be determined in the Phase II effort as each maintenance and platform have different systems.

8. Naval aviation platforms to include P-8, MH-60, unmanned systems, etc.
4/20/25  Q.
  1. What specific types and formats of heterogeneous data sources should the virtual agent prioritize for integration (e.g., radar, sonar, text-based intelligence reports, or environmental data)? Are there particular data streams or formats (e.g., structured vs. unstructured, real-time vs. batch) that are critical to Navy operations, and what are their typical volumes or update frequencies?
  2. Can you clarify the expected level of explainability for the virtual agent’s outputs? Does the Navy require detailed causal explanations for all insights, or is a combination of high-level summaries and detailed reasoning for critical decisions sufficient?
  3. What are the key performance metrics for evaluating the virtual agent’s effectiveness in Phase I and Phase II? Are there specific benchmarks for accuracy, processing speed, or robustness in dynamic maritime scenarios, and how should trade-offs (e.g., speed vs. accuracy) be prioritized?
  4. Are there existing Navy systems or platforms with which the virtual agent must integrate, and what are the associated interface or security requirements? Does the agent need to interface with specific command-and-control systems or data management platforms, and are there mandatory cybersecurity protocols (e.g., DoD standards)?
  5. What are the operational constraints for deploying the virtual agent in real-time maritime environments? Are there limitations on computational resources (e.g., edge vs. cloud processing), power consumption, or latency that the system must adhere to for practical use?
   A. 1. A mixture of both structured and unstructured data is envisioned. Formats are dependent on the specific source of such data and the platform with which it is originating from (manned versus unmanned). Update rates are also constrained by this data transfer between platforms and architectures.
Environmental data would be from on-board (air temperature, winds, etc.) or off-board sensors (AN/SSQ-36 (sound speed profiler), buoy data, etc.
Aircraft sensor data is based on current implementation and for the most part real-time with playback capability.

2. Making the options behind the decision(s) and prediction(s) (forecast) provided be described by reasoning that has been applied. Clear and concise explanations which build trust (repetitive similar solutions provided without input variation) to the operator.

3. Phase I. Is based on the approach, architecture and implementation as described in the Phase I proposal. You will be evaluated on the relationship between the outcomes suggested in the proposal and performed in Phase I.
Phase II. The system should not degrade current performance metrics, increase operator’s cognitive workload, misinform decision-making (hallucination). Processing time should be faster than human performing the same function. They are prioritized equally (speed and accuracy).

4. Phase I. All work is unclassified.
Phase II. Will consider various implementation schemes based on work being described in Phase I.

5. Phase I. Prototype system description can be either stand-alone or cloud based.
Phase II. System will not be cloud accessible to comply with EMCON conditions. System will operate solely on organic platform compute capacity or augmented based on provided rational from Phase I.
4/8/25  Q. We are very interested in Topic N252-089 and would appreciate your insight on the following:
  1. Would a neurosymbolic AI agent using Azure OpenAI + knowledge graphs be aligned with the Navy’s expectations?
  2. Can semantic reasoning via Microsoft Copilot Studio help unify diverse datasets in operational contexts?
  3. What are the preferred standards (e.g., OWL, RDF, SPARQL) or ontologies to align with Navy data environments?
  4. How critical is natural language generation (NLG) vs. visualization for operator-facing interfaces?
  5. Are there current systems or platforms you expect Phase I solutions to integrate with or augment?
   A.
  1. The goal of the SBIR is to be enlightened by various implementation and tools available and not be narrowly focused.
  2. The goal of the SBIR is to be enlightened by various implementation and tools available and not be narrowly focused.
  3. The envisioned system is to operate/integrate over multiple platforms both current and future. Similar general response to Q1 & Q2. An modular open systems architecture (MOSA) compliant approach is a direction to head towards.
  4. The Phase I concept phase is for the small business to the provide the Government with an implementation design concept. A pros/cons methodology could be utilized to explain the chosen path.
  5. Given this is a mainly an outline/approach with some substantiation of methodology a generic platform and data streams is envisioned


[ Return ]