DON27TZ02-NV011 TITLE: Heterogenous Swarm Control through Super Intelligence-embodied Architecture

OUSW (R&E) CRITICAL TECHNOLOGY AREA(S): Applied Artificial Intelligence (AAI)

COMPONENT TECHNOLOGY PRIORITY AREA(S): Advanced Computing and Software;Human-Machine Interfaces;Trusted AI and Autonomy

PROJECTED CMMC LEVEL REQUIREMENT: Level 2 (Self)

OBJECTIVE: Develop Agentic Super Intelligence (SI) architecture and associated applications for control of low-cost Commercial off-the-shelf (COTS) uncrewed systems (Group 1 unmanned aerial vehicle [UAV] or small unmanned ground vehicle [UGV]) for use in adaptable heterogenous swarming applications.

DESCRIPTION: Current methods of heterogeneous swarming control often involve individual entities with incompatible messaging protocols, ridged data interpretation, and lack of broader situational awareness, generating a protocol/perception problem that results in fragmented decision-making. Model Context Protocol (MCP), Large Language Models (LLMs), and related Agentic SI architectures, can enable a Universal Serial Bus Type C (USB-C) solution to this fragmentation problem by allowing individual entities to communicate with human operators or other entities through natural language.

The Department of the Navy is seeking development of Agentic SI-enabled COTS small UAVs or small UGVs USB-C capabilities for demonstration of applicability in Naval and Marine Corps tasks. The goal is to deliver a small heterogenous swarm of low-cost uncrewed systems capable of receiving tasks through a natural language interface and performing complex tasks such as inspection of a facility for damage, evaluating road conditions, conducting security operations, and/or intercepting of other platforms. Performers should focus on software development to run on low cost and edge compute devices attached to highly expendable COTS platforms. Submissions will leverage preexisting DoW compliant hardware deployments using either compute resources organic to the UAV/UGV platform or flight-ready modules (e.g., Single Board Computer NVIDIA JETSON NDAA) with favorable Size, Weight, and Power (SWAP). Submissions should identify the unique nature of their Agentic-SI capability in accomplishing the objective.

PHASE I: Document and demonstrate an approach for Agentic SI control of small UAV or UGV. Required deliverables for Phase I include a demonstration and design approach for validation. GNSS may be used at this stage, but overall architecture must be GNSS optional. Hardware used in Phase I needs to be either NDAA complaint or have NDAA compliant analogs listed for Phase II. Individual unmanned systems must be tasked by human or other agent through natural language and respond similarly.

Demonstrate at least one Group 1 UAV or small UGV (such as an RC Car) performing 3 of the following 5 tasks. No priority is given to this list.

• Go to user defined location with a maximum flight distance of 2km
• Report scene observations from on-board sensors (e.g., individual vehicles including class and color, people in the area, structures including temporary)
• Report on user specified object in scene (e.g., confirm presence of user specified "Red Van", or "Fishing Boat"
• Follow internal reference frame "left, right, up, down, turn left, move right 10 meters" commands and/or global reference
• Follow a leader vehicle or person

PHASE II: Using approach and results developed in Phase I, build and deliver 10-20 prototype UxSs and software that can demonstrate all of the following tasks.

Tasks

• Inspection of buildings, airfields, or roadways after weather events for debris or damage
• Go to location and report on activities in the area
• Interception of UxS
• Logistics resupply
• Security Alerts under GNSS Denied Conditions

Prepare a Phase III commercialization/transition plan.

The swarm must be composed of at least 3 different hardware configurations and require minimal user configuration. This can be accomplished by adding different sensor packages to the same base platform or altering base platform to demonstrate different capabilities. A single user must be able to command and control the entire swarm from single device using bidirectional natural language. The swarm must decide how many and of which entities are needed to complete the task. By the end of Phase II, the entire swarm must be capable of operating without internet connectivity or GNSS. Reduced capabilities in GNSS-denied environments are acceptable.

PHASE III DUAL USE APPLICATIONS: Successful completion of Phase II results in products immediately useful to Heterogenous Swarming efforts for the Department of the Navy. Additional support would include field support, continued technical refinement, technical documentation, and integration into naval systems.

Additional commercial and government applications exist across several commercial sectors such as disaster relief, agriculture, and security applications in remote areas. SI enabled UAVs to aid post-disaster search-and-rescue teams by detecting hazards on roadways, locating survivors, delivery lifesaving supplies, and automatically coordinating with other SI agents to ensure area coverage. Agricultural applications include remote crop inspections and dynamic spraying based on real-time feedback. Lastly, SI enabled UAVs can be used as part of a security or building inspection posture for remote or hard to surveil areas by responding to remote sensors and confirming presence of absence of threats.

REFERENCES:

1. Cardenas, I. S., Arnett, M. A., Yeo, N. C., Sah, L. and Kim, J.-H. "(2026, March 27). "osClaw: An OpenClaw ROS 2 Framework for Agentic Robot Control and Interaction, March 27, 2026." doi:https://doi.org/10.48550/arXiv.2603.26997
2. Chhetri, G. et al. "Model Context Protocols in Adaptive Transport Systems: A Survey." ACM Comput Surv, 1(1), August 2025. https://doi.org/10.48550/arXiv.2508.19239
3. Li, H., Cao, X., Zeng, Z., Wu, Y., Zhang, Y. and Xia, Y. "OpenGo: An OpenClaw-Based Robotic Dog with Real-Time Skill Switching." arXiv, April 3, 2026. https://doi.org/10.48550/arXiv.2604.01708
4. Ramos-Silva, J. N. and Burke, P. J. "A Universal Large Language Model -- Drone Command and Control Interface." arXiv, May 6, 2026. Retrieved from https://doi.org/10.48550/arXiv.2601.15486

KEYWORDS: Generative SI; Large Language Models; LLM; Model Context Protocol; MCP; Modularity; Unmanned Aerial System; UAS; Unmanned Ground Vehicle; UGV; Unmanned Aerial Vehicle; UAV

TPOC 1: Peter Squire
peter.n.squire.civ@us.navy.mil

TPOC 2: Meredith Horne
meredith.e.horne.civ@us.navy.mil