FPV Terminal Targeting AI

The FPV Terminal Targeting AI is a computer-vision module built by Abto Software, a Lviv software house whose defence practice supplies autonomy components to Ukrainian FPV drone makers rather than complete airframes. Introduced in 2023, the module addresses one specific problem that has come to define the drone war on both sides of the front line: the final seconds of a strike, once the operator’s radio link has been jammed or the video feed cuts out, and the drone must finish the intercept on its own.

The problem is now the routine case, not the exception. Russian electronic warfare — the Pole-21, Shipovnik-Aero and truck-mounted jammers that ride with most manoeuvre units — has made the 900 MHz, 1.2 GHz and 2.4 GHz bands that FPV operators use increasingly unusable at close range to the target. Ukrainian units and their software suppliers responded by pushing autonomy to the last few hundred metres: the pilot draws a bounding box over the target on the video feed, the onboard model locks on, and the drone runs the terminal phase on its own even if the link is severed.

Abto describes the module as a real-time target-lock stack running on the drone’s onboard compute. Systems in this category typically use an NVIDIA Jetson-class module — the Jetson Nano and Orin Nano have become the workhorses of Ukrainian terminal-guidance kits — paired with a lightweight object-detection network and a tracker that hands the drone off from operator control to autonomous descent once lock is established. The base FPV airframe itself commonly runs Betaflight or iNav ; the CV module sits above the flight controller, feeding target-relative corrections rather than replacing it. Neither GPS nor a live datalink is required in the terminal phase, which is the whole point of the design.

Ukraine is the module’s public operator. Abto has not published unit counts, and the product has not surfaced by name in Ukrainian government procurement announcements the way whole-drone systems from the larger FPV manufacturers have. Software-module vendors in this segment tend to operate one layer below the visible market, licensing autonomy stacks to the many small Ukrainian drone assemblers rather than fielding branded systems of their own.

The category is now crowded. Auterion’s Skynode S carries a comparable terminal-lock capability, and several Ukrainian teams — including Swarmer’s swarm-coordination stack — push in the same direction. What distinguishes vendors here is less any single feature than the speed at which they can retrain their models against new Russian camouflage, decoys and countermeasures as those appear on the front line.