Swedish startup Helsing demos autonomous attack drones powered by NVIDIA Jetson Orin Nano that independently identify and strike targets with no human input or external comms.
Scaleout Systems, a Swedish AI startup, demonstrated an autonomous loitering munition that independently identified and attacked targets using onboard artificial intelligence. In a recent test conducted as part of BAE Systems Bofors' Winter Demo 2026, the drone—a low-cost munition developed under the Affordable Loitering Modular Ammunition (ALMA) program—detected and geolocated potential targets, ranked an armored engineering vehicle as the highest-priority threat, autonomously flew to it, and dropped an explosive. The entire mission, from deployment to impact, took under 320 seconds, with approximately 200 seconds spent on reconnaissance. Beyond a button press to initiate the system, manual pilot input remained optional; the designated pilot served only as a failsafe controller. The mission required no external communications, demonstrating resilience against electronic warfare.
All processing ran onboard the airframe. Scaleout's demonstration video, "Technical Demo: Onboard Edge Intelligence for Autonomous UAV Missions," shows the drone identifying potential threats with vision-based AI while the designated pilot maintained optional manual override. The munition is more accurately termed a loitering munition than a kamikaze drone, as most of its flight time is spent searching, ranking targets, and waiting to strike.
Scaleout combines its Edge platform with federated learning to create a Tactical Computer Vision Network (TCVN), in which devices train AI locally and share model updates for rapid adaptation without centralized processing. In a separate demonstration conducted in February at BTC Karlskoga, Sweden, an Airolit S1 airframe ran the YOLOv8 Nano object-detection model on Nvidia's Jetson Orin Nano at minus 18 degrees Celsius. The system achieved 30 frames per second at 20 meters per second airspeed, with target latency of 30 milliseconds or less, sustained in field conditions. Ranging without a depth sensor was possible using a pinhole camera model combined with known object dimensions.
A June 2026 demonstration tested Scaleout Edge's resilience during degraded communications. Two ground nodes operated under a cloud-hosted Scaleout Edge control plane: ALPHA, a forward-deployed node at the air base with stable connectivity, and BRAVO, a lab node in Uppsala whose connection was first degraded and then severed entirely. BRAVO maintained inference and active learning at full frame rate while offline, logging detections locally, then backfilled them on reconnect in priority order for heartbeat, critical alerts, drift detection, model updates, and telemetry. This test mimicked the impact of external interference and jamming.
Scaleout's FEDAIR project is part of NATO's DIANA accelerator program, receiving 100,000 euros in development funding, training, and test access. These demonstrations and funding rounds precede any official NATO procurement order or combat deployment. A UN expert group on lethal autonomous weapons recently concluded that human judgment and control remain required to comply with the laws of war; the group's recommendations will be reviewed in Geneva in November. Nevertheless, Scaleout's demonstrations establish that autonomous target selection can run entirely on affordable, commercially available processors aboard the munition itself.