Tactical ISR Processing, Exploitation and Dissemination (PED): Adaptive AI and Autonomy at the Tactical Edge

The National Geospatial-Intelligence Agency (NGA), through its Rapid Capabilities Office (QX), in coordination with ONI and ACC-RI, is seeking prototype solutions to enable sustained tactical intelligence, surveillance, and reconnaissance (ISR) operations at the edge through AI-enabled computer vision capabilities. This prototype effort addresses a critical capability gap in contested and austere communications environments where warfighters depend on continuous custody of objects of interest for mission success and national security decision-making. Current computer vision models degrade when encountering new environmental conditions and require reach-back to enterprise cloud infrastructure for retraining, a dependency that challenging operational conditions preclude. This challenge solicits novel prototype solutions under an Other Transaction Authority (OTA) vehicle to rapidly develop, test, and validate edge-deployable computer vision capabilities for unclassified small unmanned aerial systems operating in low-connectivity environments. The prototype will focus on object detection, tracking, and re-identification in mission-relevant data environments, with emphasis on low-shot learning approaches that enable operators and analysts to deploy and adapt capabilities at the point of need with minimal labeled data and minimal coding expertise. Integration with standard GEOINT data formats, tactical data dissemination systems including the TAK ecosystem, and operator display environments is required to ensure detections reach end users in usable, interoperable formats that support continuous custody of objects critical to tactical intelligence operations.
## Problem Statement
Tactical sensors increasingly hand off custody of objects of interest to computer-vision-based tracking, but current models degrade in new environmental conditions and require reach-back to enterprise cloud infrastructure for retraining, which challenging conditions preclude. Analysts and operators need a capability that can be deployed and adapted at the point of need, with minimal labeled data and minimal coding expertise, to maintain continuous custody of objects critical to tactical intelligence and national security decision-making.
## Overall Objective
Prototype, test, and validate an edge-deployable computer vision capability for object detection, tracking, and re-identification in mission-relevant data environments that enables warfighters to maintain continuous custody of objects of interest in contested and austere communications environments. The prototype will demonstrate low-shot learning approaches that allow operators and analysts to deploy and adapt computer vision models at the point of need with minimal labeled data and minimal coding expertise, operating on size, weight, and power (SWaP)-constrained edge hardware without cloud dependency. The effort will validate performance against defined F1, precision, and recall thresholds while maintaining custody across multiple simultaneous detections and re-identifying objects of interest following loss of custody. Integration with standard GEOINT data formats, ontology standards, tactical data dissemination systems including the TAK ecosystem, and REST API architectures will ensure detections reach analysts and operators in usable, interoperable formats that support tactical intelligence and national security decision-making. The prototype will undergo operational evaluation to inform future development and transition decisions, with documented outcomes providing a foundation for follow-on production integration and scaling.