Small AI models let drones autonomously identify and attack battlefield targets

Scaleout Systems, a Swedish startup started by researchers from Uppsala University in 2018, has adapted its edge-focused machine learning technology for military use following Russia’s 2022 invasion of Ukraine. Rather than relying on large, centralized AI models from major cloud providers, the company emphasizes compact computer-vision models that can run on the constrained hardware found in small drones, pilot tablets and forward-deployed workstations. The company was selected for NATO’s Defence Innovator Accelerator for the North Atlantic (DIANA) Challenge Program in 2025 and has worked on a Federated Aerial Intelligence for Recon effort to tailor models to edge computing hardware. Scaleout’s approach uses intermittent, selective updates between local devices and platoon- or company-level computing nodes; those nodes aggregate battlefield data to retrain models and then distribute improved versions back to edge devices when connectivity allows. Scaleout demonstrated the approach publicly through exercises and projects including the ALMA program led by BAE Systems Bofors. At a Winter Demo 2026 event in Sweden, an ALMA loitering munition used onboard AI to detect, identify, geolocate and prioritize potential threats, then autonomously flew to and struck a mission-defined high-value target — an armored engineering vehicle — while retaining the option for human control. The company also ran tests at a Swedish Air Force base in Uppsala in June that showed a forward-deployed computing node could continue inference and active learning after losing contact with a central lab node; once connectivity returned, the local model updates were shared. Scaleout says this federated learning strategy could allow AI models to continuously improve from distributed sensor data and potentially be scaled across allied forces and regions.