Army CRADA Puts Ukraine-Hardened Drone AI on Dual Tracks

Safe Pro Group signed a Cooperative Research and Development Agreement with Army ERDC on August 12 to feed its SPOTD computer vision into uncrewed aerial system data for operational, disaster and mission-critical analysis. The same platform has already processed nearly three million drone images in Ukraine and flagged more than 54,000 explosive threats.

The deal is not a one-off sensor buy. It routes battle-tested edge AI into both warfighter planning and civil geospatial work under the same roof.

That pairing matters because the imagery problem is shared. Small units, engineers and disaster teams all need plotted objects and terrain products faster than manual review can deliver. The CRADA gives that shared need a formal research channel inside Army engineering science.

What the CRADA Locks In

The project title is Unified Situational Intelligence: UAS Enabled Data for Operational, Disaster, and Mission Critical Analysis. It pairs Safe Pro with the United States Army Engineer Research and Development Center, the research arm of the Army Corps of Engineers.

CRADAs let federal labs share people, facilities and know-how with non-federal partners without handing over cash the other way. Safe Pro brings its patented object-detection stack; ERDC brings engineering geospatial expertise and a path into Army and civil-authority use cases. That structure is standard under the Technology Transfer Act, as laid out in how Army ERDC CRADAs exchange expertise.

We are proud to expand our growing engagement with the U.S. Army to now include the collaboration on applying our patented AI-powered technologies to not only improving the effectiveness of military operations, but to potentially a wide array of civil works, military engineering or disaster response applications.

Dan Erdberg, chairman and CEO of Safe Pro Group, said that in the company announcement.

The emphasis sits on dual-use products: analytic outputs from drone and sensor imagery that serve both the small-unit warfighter and civil authorities who manage hazards, infrastructure and disasters.

Because no government cash flows to the partner under a CRADA, progress depends on each side putting real assets on the table. Safe Pro’s contribution is the detection software and the Ukraine-trained models. ERDC’s contribution is geospatial engineering depth and access to the mission problems those models must solve next.

How SPOTD Turns Raw Video Into Shareable Maps

SPOTD stands for Safe Pro Object Threat Detection. It runs AI computer vision on imagery and video from virtually any drone, then converts the raw feed into high-resolution 2D and 3D geospatial models that can be shared quickly.

• Detects and classifies more than 150 types of landmines and unexploded ordnance
• Builds orthomosaics, digital surface models, vegetation height and terrain slope layers
• Supports automated route planning once threats are plotted
• Works cloud-side (AWS-powered SpotlightAI) or on the edge without connectivity
• Processes commercial off-the-shelf drones as well as Army Short-Range Reconnaissance platforms

The software was built to replace slow manual review of thousands of overlapping photos. Human analysts still set the mission and verify results, but the first pass now runs at machine speed.

The output stack is what makes the detections useful beyond a single screen. Orthomosaics and surface models give planners a common picture. Vegetation height and slope layers add movement context. Once threats are plotted, route planning can run against that same map instead of a separate workflow.

The Ukraine Dataset That Trained the Model

SPOTD has operated in active environments in Ukraine for nearly three years. Humanitarian mine-action groups and government agencies upload drone surveys; the system returns GPS-tagged detections.

Current public tallies on Safe Pro’s dashboard show the scale:

• 2,982,446 images processed
• 54,328 landmine and UXO detections
• 15,344 hectares surveyed (about 37,900 acres)
• 42 terabytes of data analyzed

Those real-world Ukraine detection tallies form one of the largest proprietary small-threat imagery sets available. Earlier milestones tracked the climb from roughly 23,000 detections to more than 50,000 as coverage expanded.

Norwegian People’s Aid used the AWS-hosted version of the platform and reported it could cut survey analysis times by nearly 90 percent versus manual review, with estimated cost cuts around 80 percent and per-image analysis dropping to roughly one-fifth of a second. That speed matters when contaminated farmland and civilian areas stretch across tens of thousands of square kilometers.

The climb from roughly 23,000 detections to more than 50,000 also shows how the corpus compounds. Each new survey adds labeled examples. Those labels feed the same models that later run on edge kits and Army Short-Range Reconnaissance feeds.

NODE-X Puts the Same AI in a Soldier’s Pack

Cloud processing works for deliberate surveys. Maneuver units need answers in minutes without a backhaul link. Safe Pro’s answer is NODE-X, a backpack kit with real-time edge inference, ruggedized GPU laptops and OnSight software that pairs with Army-approved Short-Range Reconnaissance drones.

In April 2026 the company completed a live minefield exercise with NODE-X. Drone imagery of live scattered mines was processed on the edge; detections and supporting 3D products reached commanders fast enough to support mission planning. Additional Army exercises were planned through the second quarter of 2026.

The training data for that edge stack is the same Ukraine corpus: millions of images and tens of thousands of confirmed small threats. The company has also integrated the detection layer onto platforms shortlisted for the Army SRR program of record, claiming sub-second analysis per image for more than 150 threat types.

Cloud and edge are two deployment modes of one detection layer, not separate products with separate libraries.

Civil Works and Disaster Response Ride the Same Pipe

ERDC’s mission covers national security, economic infrastructure and disaster risk reduction. The CRADA language deliberately keeps civil authorities in the frame. The same object detection that flags a TM-62 antitank mine can mark flood debris, damaged bridges, hazardous spill markers or other engineering hazards once the models are extended.

That dual-use path is the second-order payoff. Military funding and soldier feedback harden the edge kit and the detection models. Civil agencies and humanitarian groups inherit faster survey tools without building their own computer-vision stacks from scratch. The AWS case with Norwegian People’s Aid already showed the humanitarian side of the ledger in Ukraine; the ERDC agreement opens a formal channel inside US Army engineering science.

The mechanism is straightforward. Threat classes and hazard classes both begin as objects in imagery. Once the pipeline can detect, locate and publish them as shareable geospatial layers, swapping the object library is an extension problem rather than a ground-up rebuild.

Speed Gains Come With Familiar Risks

Units that once waited hours or days for imagery analysts can now get plotted threats and route options while still on the objective. Demining organizations free scarce human experts for verification and clearance instead of first-pass photo review. Commanders gain shared 3D products that travel across echelons.

The risks are the ones every automated detection system carries. False positives can slow a breach or send sappers to empty ground. False negatives are worse. Edge systems reduce bandwidth dependence but still need power, training and clear tactics for when the AI and the soldier disagree. Broader debates about military AI reliability, including over-trust in machine recommendations, apply here even if this specific CRADA has drawn little independent public pushback so far.

What we know

• CRADA signed August 12, 2026 with Army ERDC for UAS-enabled AI analysis
• SPOTD already fielded nearly three years in Ukraine with public detection counts above 54,000
• NODE-X demonstrated live mine detection on the edge in an Army exercise
• Dual-use language covers warfighter, civil works, military engineering and disaster response

What remains open

• Exact deliverables, timelines and any follow-on procurement under the CRADA
• Measured false-positive and false-negative rates on Army-specific threat libraries
• Scale of eventual fielding across SRR fleets or other UAS programs

Drone security itself stays contested. Recent earlier drone-security claims pattern shows how quickly attribution and capability claims circulate around uncrewed systems; the Safe Pro stack focuses on the sensor-to-map layer rather than the flight-control plane, yet operators still treat the whole kill chain as a single attack surface.

How the Army Touchpoints Line Up

The ERDC agreement did not arrive in isolation. Safe Pro has built a sequence of Army-facing milestones that share the same detection stack and the same Ukraine-derived training base.

1. Nearly three years in Ukraine – operational surveys that produced the public image and detection tallies now feeding the models.
2. SRR integration work – detection layer placed on platforms shortlisted for the Army Short-Range Reconnaissance program of record.
3. April 2026 – live minefield exercise with NODE-X edge processing and rapid commander products.
4. Through second quarter 2026 – additional Army exercises planned after the April event.
5. Project Convergence demonstrations – further Army touchpoints cited alongside the exercise lane.
6. August 12 – CRADA with Army ERDC for Unified Situational Intelligence.
7. Late August 2026 – Innovation Day with live drone operations scheduled.

Each step tests a different slice of the same pipe: data quality, edge timing, platform fit, then formal research partnership. The CRADA is the piece that ties warfighter and civil-works requirements into one lab relationship.

Where the Next Months Point

Safe Pro has stacked multiple Army touchpoints: SRR integration work, the live minefield exercise, Project Convergence demonstrations and now the ERDC CRADA. An Innovation Day with live drone operations was scheduled for late August 2026. Revenue figures released around the same period showed sharp growth driven by government AI contracts, though those numbers sit outside the technical scope of the CRADA itself.

The practical test is whether NODE-X and SPOTD products move from experimental lanes into routine small-unit and engineer use, and whether civil agencies pick up the same analytic products for disaster and infrastructure missions. The dataset and the dual-use framing already exist. The CRADA gives both sides a formal table to refine them together.

Open questions on deliverables, error rates and fielding scale will decide how far that table reaches. Until those answers land, the public record still rests on the Ukraine tallies, the NODE-X exercise result and the dual-use language written into the agreement.

Frequently Asked Questions

What is Safe Pro’s SPOTD platform?

SPOTD is patented AI computer vision software that analyzes drone imagery and video to detect and classify explosive threats and other objects, then generates 2D and 3D geospatial models for sharing. It runs in the cloud or on edge hardware and works with commercial and military drones.

How many types of landmines can the system identify?

Safe Pro states the platform can identify more than 150 types of landmines and unexploded ordnance, drawing on real detections that include common antipersonnel and antitank mines as well as rockets and cluster munitions seen in Ukraine.

What does a CRADA with the Army ERDC involve?

A Cooperative Research and Development Agreement lets the Engineer Research and Development Center share personnel, facilities and technical expertise with a non-federal partner such as Safe Pro. The government does not fund the partner under a CRADA; both sides contribute resources toward a defined research project that fits the lab’s mission.

How large is the Ukraine training dataset behind SPOTD?

Public figures list roughly 2.98 million images processed, more than 54,000 confirmed threat detections and coverage of about 37,900 acres. The dataset continues to grow as new surveys are analyzed.

How long has the technology been used in Ukraine?

Safe Pro reports nearly three years of operational deployment in Ukraine supporting humanitarian mine-action organizations and government agencies with AI-assisted drone survey analysis.