By 
The National Geospatial-Intelligence Agency (NGA) and the Army’s 18th Airborne Corps in May successfully tested a prototype tool that uses imagery data from small drones and can train artificial intelligence models in the field during operations to detect anomalous patterns, targets, and objects and give warfighters and other users greater situational awareness at the tactical edge.
The GEOINT AI/ML Based Light-Edge Resilient (GAMBLER) system will provide the AI-based detections into portable devices carried by warfighters, including in disconnected environments, NGA said on Wednesday. The detection models may already exist, or be trained on the fly in the field as warfighters and other users label various objects observed by the drones as part of routine collection, processing and dissemination of data.
As the models are trained and new detections are automatically made, GAMBLER will alert the various users and reduce time on target, NGA said. The detections and patterns could include vehicle types, individuals, and other anomalous activity.
“GAMBLER is designed to be operated by the warfighter,” the chief engineer of NGA’s Warfighter Support Office said in a statement. “It’s about pairing the capabilities of GEOINT, machine learning, AI and mesh networks to reduce the amount of time it takes to identify important objects and quickly pass that tactical situational awareness along to where a decision can be made or action taken.”
A beta version of GAMBLER the was used in the 18th Airborne Corps’ Scarlett Dragon exercise in North Carolina in early May. Parsons Corp. [PSN] is the prime contractor for the system.
Oak Ridge National Laboratory’s MAPSTER software is a key ingredient of GAMBLER, cataloging and storing the data provided by the small drones. MAPSTER delivers data to “the GAMBLER system, where the AI models live, to run against the data, enrich it with AI, and disseminate it out to the end user,” an NGA official said in a video about the system posted on YouTube.
For the tests, imagery data was collected, models were built in real time, data was processed, and disseminated to multiple systems, NGA said. In addition to installing GAMBLER software on low-cost handheld devices, the testers successfully allowed detections and images to be sent through portable radios to locations throughout the U.S., the agency said.
NGA’s premier AI-based detection system is called Maven, an enterprise system in use by the Combatant Commands and other users. Maven is based on enormous amounts of data that can take months to train AI models. GAMBLER is aimed at training models in hours, days, or weeks.
