AI Governance
Data Centers
Drones
Climate
Privacy
2026 Maine Senate
Bob Dylan
Scaleout’s Military Drone AI
Coverage from DroneXL, Tom's Hardware, and others
Scaleout Systems’ edge AI has been demonstrated with BAE Systems Bofors’ ALMA loitering munition, using onboard computing to detect, identify, and geolocate a target without depending on an external server.
Tests also showed inference continuing after communications were lost. The related FEDAIR approach shares model updates rather than sensor footage, aiming to support distributed learning while limiting data transfer. These demonstrations raise important questions about how human oversight, vetoes, and responsibility should work when a system can select or engage targets; they do not establish combat deployment or a NATO procurement decision.
If you read one thing
It clearly introduces the demonstrated targeting capability, communications resilience, and unresolved human-control questions.
The evidence
It adds concrete reported performance figures and clarifies that the demonstrations had not resulted in procurement or combat deployment.
Best explainer
It explains how compact edge models and federated learning support drone operation when battlefield communications are unreliable.
Autonomous target selection and engagement is demonstrated
Scaleout’s ALMA system detected, ranked, selected, and engaged an armored vehicle without a human selecting the specific target or issuing continuous commands. The demonstrations establish perception-to-attack execution as a tested capability, though not yet a deployed one.
Edge autonomy is designed for communications-denied conditions
The system’s onboard AI continued inference, logged detections, and conducted active learning after communications were lost, then synchronized after reconnection. Link-loss resilience and decentralized model operation are central features for jamming-contested environments.
Human-control and accountability requirements remain unresolved
The demonstrations leave open where operator veto or abort authority exists after a drone loses communications and who is responsible when autonomous targeting is wrong. Existing policy statements and human-oversight expectations do not resolve this specific lost-link accountability gap.
approximately 200 seconds
reconnaissance duration
“The system spent approximately 200 seconds conducting reconnaissance and completed the full mission in under 320 seconds without requiring communications, supporting Scaleout’s claim that it can operate in environments affected by electronic warfare.”
under 320 seconds
full mission duration
“The system spent approximately 200 seconds conducting reconnaissance and completed the full mission in under 320 seconds without requiring communications, supporting Scaleout’s claim that it can operate in environments affected by electronic warfare.”
-18 degrees Celsius
operating temperature
“In a separate February demonstration at BTC Karlskoga in Sweden, an Airolit S1 airframe operated in snow and temperatures of -18 degrees Celsius. It ran the YOLOv8 Nano object-detection model on Nvidia’s Jetson Orin Nano. Scaleout reported approximately 30 frames per second at around 20 meters per second, with target latency of 30 milliseconds or less.”
around 20 meters per second
airframe speed
“In a separate February demonstration at BTC Karlskoga in Sweden, an Airolit S1 airframe operated in snow and temperatures of -18 degrees Celsius. It ran the YOLOv8 Nano object-detection model on Nvidia’s Jetson Orin Nano. Scaleout reported approximately 30 frames per second at around 20 meters per second, with target latency of 30 milliseconds or less.”
30 milliseconds or less
target latency
“In a separate February demonstration at BTC Karlskoga in Sweden, an Airolit S1 airframe operated in snow and temperatures of -18 degrees Celsius. It ran the YOLOv8 Nano object-detection model on Nvidia’s Jetson Orin Nano. Scaleout reported approximately 30 frames per second at around 20 meters per second, with target latency of 30 milliseconds or less.”
No new member articles were supplied, so there is no evidence of a material real-world change since the prior state.
Previously
Swedish defense demonstrations show military drones using onboard AI to identify, prioritize, and engage targets while continuing local inference through communications disruption; procurement and human-control arrangements remain unsettled.