NATO-Backed Startup Enables Autonomous Offline AI Drone Target Recognition and Strikes
Swedish defense startup Scaleout Systems, backed by NATO's DIANA accelerator, has demonstrated lightweight edge AI models running directly on drone hardware to autonomously detect, identify, and strike battlefield targets without requiring a central network connection. As part of BAE Systems Bofors' ALMA project, the system processed all target data onboard and executed strike missions even during complete communication blackouts. Modern battlefields suffer from heavy electronic warfare and communications signal jamming, rendering cloud-dependent autonomous military systems unusable. By moving computer vision inference to local edge devices and utilizing federated learning, military units can maintain tactical strike capabilities while updating AI models without transmitting sensitive raw sensor data. Scaleout avoids large cloud-based models in favor of lightweight computer vision models built for embedded chips, operator tablets, and mobile field command nodes. Through a federated learning architecture, localized command posts retrain models using fresh battlefield data and sync updated weights across devices when connectivity is intermittently restored.
## BACKGROUND
Edge AI refers to processing machine learning algorithms directly on local hardware devices rather than relying on remote data centers, drastically reducing latency and working offline. Federated learning is a decentralized machine learning technique where devices train local models on onboard data and share only model updates (weights) instead of raw data. NATO's DIANA (Defence Innovation Accelerator for the North Atlantic) is an acceleration program designed to foster dual-use defense and security technologies across member nations.