Scaleout Demonstrates Drone and Field AI Built for Disrupted Links
The Swedish startup’s decentralized system shares model updates instead of raw sensor data, while its ALMA work shows onboard AI prioritizing and engaging a target under retained operator control.
Listen to this story
The audio brief
Story brief
3 key pointsScaleout Systems demonstrated a defense AI stack designed to keep adapting when communications fail. In January, its onboard system helped a drone detect, geolocate, prioritize, and engage an armored engineering vehicle during Sweden’s ALMA trial; in June, a forward node at Uppsala continued inference and active learning after losing its laboratory connection, then synchronized updates after reconnection.
- 01
The June Uppsala exercise tested model continuity during a live link outage, not merely offline inference.
- 02
Devices share selected model updates with a local node instead of continuously transmitting raw sensor data.
- 03
The ALMA drone completed target prioritization and flight without direct human commands; an operator could still intervene.
Scaleout Systems is making a case that battlefield AI need not go offline when communications do. The evidence so far is a pair of Swedish demonstrations: a field computer that kept learning after its central link failed, and a drone that identified, prioritized and attacked a target with onboard AI.
Through its Federated Aerial Intelligence for Recon project, Scaleout is adapting machine-learning models for drones, pilot tablets and field command posts. The work followed the company’s 2025 selection for NATO’s DIANA Challenge Program, which supports defense-focused innovation.
Training at the edge, not a distant server
The system keeps inference, the step where a model produces an output, on a drone or nearby field hardware. Rather than continuously sending sensitive raw sensor data, devices share selected model updates with a local computing node. That node can retrain the models using information from several devices and send updated versions back when a connection is available.
The aim is to let local systems adapt as conditions change without relying on continuous contact with a central server. Scaleout’s chief executive, Andreas Hellander, told Ars Technica that models trained in one setting may not perform as well in another.
The ALMA trial put target selection onboard
Scaleout is participating in the Affordable Loitering Modular Ammunition project led by BAE Systems Bofors. At a January demonstration in Sweden, a drone used onboard AI to detect, identify and geolocate potential threats. It then prioritized an armored engineering vehicle under its mission, flew to it and dropped an explosive.
A human operator could still control or direct the drone. But in the demonstration, the aircraft handled target prioritization and flew the mission without direct human commands.
A base exercise tested the network after an outage
In June, Scaleout tested the same decentralized approach at a Swedish Air Force base in Uppsala. A forward-deployed node continued inference and active learning after losing contact with a central node in Scaleout’s lab, then synchronized its updates after reconnection.
Scaleout said the Swedish military is licensed to use its main software platform. Founded by Uppsala University researchers in 2018, the company shifted toward defense applications after Russia’s full-scale invasion of Ukraine in 2022.
Sources
- arstechnica.comNATO-backed startup adapts AI for autonomous drone recon and attack missions
Reader comments
Newest comments first. Replies stay oldest first.