Onboard Detection Replaces Video Streaming

Running detection on the drone itself is what makes Ghost Dragon usable in a jammed battlefield instead of only in a clean demo. The aircraft does the heavy work in the air, on its onboard machine vision computer, then sends back small packets like target coordinates and labels instead of a constant video feed. That cuts radio load, lets the operator stay concealed, and keeps the system useful when GNSS and links are degraded or forced onto narrow channels.1 2 3 4

• Ghost Dragon is built around edge autonomy, not just cameras. KrattWorks lists an onboard quad core 1 GHz Arm machine vision computer, jamming resistant radio, GNSS spoofing detection, and return navigation even without GNSS and radio link. The product is designed so sensing, detection, and basic decision making stay on the aircraft.2 3
• This changes the data flow. Instead of pushing full thermal and RGB video over a fragile link, the drone can reduce the stream to georeferenced target information for ATAK or Kropyva style workflows, where a user mainly needs where the vehicle, person, or fire is, not every pixel captured along the way.2 4
• The broader defense pattern is toward edge processing under poor connectivity. Anduril describes the same logic across its systems, keeping computer vision and sensor fusion local at the tactical edge and avoiding dependence on distant compute. KrattWorks is applying that playbook in a much smaller, lower cost reconnaissance drone sold around €15K to €25K per system.1 4 5 6

The direction is toward drones that act more like flying sensors with their own brain than remote cameras on a leash. As electronic warfare keeps tightening the pipe between aircraft and operator, the winners will be the systems that can detect, classify, navigate, and hand off usable target data locally, with the radio link becoming a thin reporting channel instead of the product's core dependency.1 2 4 5