Ukrainian company NORDA Dynamics is preparing to release an updated version of its mass-produced Underdog auto-guidance module with artificial intelligence features, as well as the ninth version of its algorithms for fixed-wing drones. This was reported by the publication “Oboronka”, citing the company’s CEO, Nazar Bihun.
According to Bihun, the main difference between the AI-modified Underdog and the base version will be automated object detection. Previously, the operator had to find the target independently using optics and mark it with a lock-on marker; now this function will be performed by a neural network. It will recognise objects and highlight them on the screen, provided the link between the drone and the ground control unit is maintained.
The operator will be able to switch between detected objects, and for any subsequent engagement the system will require the pilot’s authorisation. The onboard neural-network model has been trained to distinguish nine separate classes of targets, NORDA Dynamics’ CEO explained.
To train the algorithms, NORDA Dynamics engineers used mainly data from real combat conditions. In particular, the training sets included datasets from Brave1 Dataroom and Avengers Labs, as well as the company’s own materials collected during field trials for specific tasks. Synthetic data accounted for only a small part of the training sample.
The updated optical auto-guidance system has also gained the ability to detect and track moving objects at a distance of up to two kilometres in favourable weather conditions.
Another development presented by the company is a navigation add-on for flights without GPS. It determines the route by combining algorithms that match optical terrain features with maps of the area and visual-inertial odometry. In addition, the system supports integration with ground-based radio beacons produced by Sine Engineering and Skyline.
According to Nazar Bihun, combining three technologies — GPS-free navigation, neural-network object recognition and auto-guidance — should form the basis for carrying out fully autonomous strike missions. He also added that NORDA Dynamics’ optical guidance system on fixed-wing drones is already capable of locking onto moving objects at a significantly greater distance than is typically provided by automatic recognition models.
At present, the company is finalising development of the pre-release AI version of Underdog and is preparing the ninth version of its auto-guidance algorithms for fixed-wing drones for an official release. The estimated timeframe for moving to fully autonomous strike missions using these technologies and optical positioning is early 2027. In addition, the developer presented an optical navigation system that makes it possible to plot a route without using GPS satellite signals.