AI systems operating in the field may not always have access to a wider network or remote computing resources. Running AI models on a drone’s onboard computer can allow it to process sensor data locally when that wider connection is unavailable.
At Winter Demo 2026 in Karlskoga, Scaleout demonstrated onboard AI as part of the wider ALMA programme led by BAE Systems Bofors. The event brought together defence companies, technology firms, academic organisations and military participants to explore different aspects of the programme.
Scaleout’s contribution focused on AI processing at the edge: detecting and prioritizing potential targets using data processed onboard the drone. The demonstration also showed the drone navigating toward the selected vehicle after an operator reviewed the AI’s selection and authorized the flight to continue.
No explosives were used, released or detonated in Scaleout’s demonstration, and no strike took place. The demonstration showed an AI and navigation workflow, not an explosive attack.
System at a glance
| Drone Platform | Airolit S1 |
|---|---|
| Onboard Compute | NVIDIA Jetson Orin Nano |
| Edge AI Orchestration | Scaleout Edge |
| Inference Engine | TensorRT-optimized YOLO-family models |
| Test Site | BTC Karlskoga, Sweden |
| Operating Temperature | −18 °C |
| Terrain | Arctic, snow-covered |
| Range Covered | 5 km |
| Target Types | TEL vehicle, command vehicle, radar vehicle |
What happened during the demonstration
The drone used onboard AI to search for and detect potential targets, then prioritize them according to the mission. An operator reviewed the system’s selection and authorized the drone to continue the demonstration flight. The operator could monitor and interrupt the flight.
After authorization, the drone navigated autonomously toward the selected vehicle, including toward its last known position when needed, and made a final approach. This demonstrated onboard processing and navigation following the operator’s approval.
The drone was not relying on the wider network for its onboard inference. It remained connected to the operator during the demonstration, so the operator could review the selection and interrupt the flight. Here, “without the wider network” does not mean “without any communication link.”
What we mean by edge AI
Edge AI means that a system processes data on or near the device that collects it, rather than depending on a remote cloud service for every computation. In this demonstration, the drone’s onboard computer ran the AI inference used to detect and prioritize potential targets.
Running inference locally is distinct from training or updating a model. A model can continue to run onboard without access to the wider network. Updating a model involves a separate process of collecting and reviewing data, validating changes and deploying an update.
Where suitable local infrastructure and connections are available, some of that work can take place at a nearby ground node instead of in a remote cloud. The specific update process and communication links depend on the system configuration.
The operator’s role
The demonstration included a human decision point. The AI processed sensor data and prioritized potential targets, while the operator reviewed the selection and authorized the flight to continue. The operator could also interrupt the demonstration.
After that authorization, the drone navigated autonomously. This describes the sequence demonstrated at Winter Demo. It should not be taken to mean that the operator authorized the drone while it had no communication path, or that the demonstration proceeded without human involvement.
The wider ALMA programme
ALMA is a broader programme led by BAE Systems Bofors. Winter Demo involved multiple organisations and areas of work. Capabilities considered within the wider programme should not automatically be attributed to Scaleout or described as part of this specific demonstration.
Scaleout’s contribution at Winter Demo was focused on onboard AI processing and the associated operator-supervised demonstration workflow.
Continuing work on edge AI
Scaleout develops software for managing AI models across distributed computing environments, including edge devices. Keeping inference close to where data is collected can reduce dependence on remote computing and may help systems continue processing information when access to the wider network is limited.
Model updates require their own review and validation process. Depending on the system, data and updates can be managed through local infrastructure, with information shared more widely when connections are available. These processes are separate from the onboard inference demonstrated at Winter Demo.
The demonstration showed how onboard AI can detect and prioritize potential targets, with an operator reviewing the system’s selection and authorizing the drone to continue before it navigated autonomously. No explosives were used, released or detonated, and no strike took place.