complete the attack after target acquisition by the operator — have become a distinct segment of weaponry at the front.

A year ago, such systems were mostly demonstrated at exhibitions; now they are entering the standard armament of individual brigades. The Ministry of Defence has officially scaled certain solutions as part of close-range air defence.

The idea is as follows: a computer vision algorithm locks onto a target and guides the drone to the point of impact, even if the operator has lost the ability to control the craft manually.

How the Technology Works in Combat Conditions

In June 2026, at Eurosatory in Paris, a Ukrainian company presented an updated version of the P1-Sun Long with a new terminal guidance system of its own design. According to the manufacturer, it locks onto a target at distances of up to one kilometre depending on the optics type, and operates on standard processors without specialised graphics accelerators.

A similar approach is used in ground-application FPV strike drones. As early as spring, participants from the GUR and the Nemesis unmanned systems brigade began procuring TFL-1 autonomy modules. This is a Ukrainian autonomy module for FPV drones. It adds computer vision and artificial intelligence capabilities to a standard FPV drone, enabling the UAV to independently complete an attack even after losing communication with the operator.

According to one GUR operator, he had previously tested dozens of drones with claimed auto-guidance, and none had shown stable results. At the same time, testing of the TFL-1 confirmed target acquisition of a moving target at a distance of 400 metres, after which the unit ordered an additional 30 modules.

To understand how communication and control of FPV drones is ensured in challenging conditions, it is worth reading the article: «FPV Repeater – How to Choose the Right One?», which covers the key aspects of building control channels. 

Automation of Close-Range Air Defence

On 8 June 2026, Mykhailo Fedorov announced that one of the Brave1 cluster participants had proven a technology that automates up to 95% of the interception cycle of a Shahed-type kamikaze drone — from launch to kinetic strike. The operator merely monitors the air situation, selects a priority target, and confirms the strike; the onboard algorithm handles the rest. The system was combat-tested in Kharkiv Oblast, and the Ministry of Defence has already decided to scale it.

According to NSDC data, approximately 100,000 interceptor drones were manufactured in Ukraine in 2025. Production capacity in this segment grew eightfold over the year, with more than 20 Ukrainian manufacturers active in the market. Meanwhile, the Commander-in-Chief of the Armed Forces of Ukraine reported that in February 2026 alone, interceptor drones destroyed more than 1,500 enemy UAVs — meaning target auto-tracking is transforming from an auxiliary function into a competitive industry.

A separate direction is developing in the form of robotic turrets with automatic guidance systems. According to the Ministry of Defence, Ukrainian remotely operated weapon stations for 7.62 mm and 12.7 mm calibre machine guns are already being used in more than ten Defence Forces units. During trials, they engaged FPV drones at distances of 50–220 metres, and the automatic recognition system detected targets at distances of up to 250 metres.

Where the Technology Is Still Imperfect

Despite developers’ claims of high recognition performance, field conditions remain the most demanding test for auto-tracking systems. The Fourth Law founder Yaroslav Azhniuk explains that the algorithm must function correctly across thousands of «edge case» scenarios — with changing lighting, in smoke and dust, with partial target occlusion, or from non-standard angles.

This is confirmed by the experience of a GUR operator with the callsign «Lincoln», who tested dozens of autonomous guidance systems before the TFL-1 appeared but did not consider any of them sufficiently effective.

Auto-tracking does not eliminate the need for a skilled operator: the algorithm holds a target well in typical scenarios but loses it during sharp manoeuvres or changes in angle, and at such moments the decision once again falls to the human.

The second systemic limitation is communications. Even the best computer vision algorithm depends on video stream quality: channel degradation caused by electronic warfare degrades the very image the AI is analysing. This is why developers increasingly position autonomous guidance not as a replacement for fibre-optic communication, but as a parallel direction — a way to preserve strike effectiveness on the «last mile» when radio communication is completely lost. This logic underpins solutions that continue target tracking purely on visual imagery, without GPS and without a radio channel, after the operator has handed control to the algorithm.

The third risk is tactical: excessive reliance on automation. Operators who grow accustomed to depending on the algorithm may lose vigilance at the very moment the system fails — and this happens more frequently in atypical scenarios: complex backgrounds, active electronic warfare, targets camouflaged as obstacles.

What’s Next?

The state is systematically expanding the foundation for developing this direction. More than 300 developments are registered on the Brave1 platform, of which around 70 AI- and computer vision-based systems are already being employed at the front.

In parallel, the Ministry of Defence, with the support of the UK government, has launched the Defense AI Center A1 — the first centre of competence for implementing artificial intelligence in defence. One of its focus areas is the integration of machine vision and AI into Ukrainian unmanned systems. The Ministry of Defence states that the strategic goal is to equip all drones used at the front with machine vision and artificial intelligence technologies.

At the same time, fully autonomous combat solutions without human involvement in the strike decision remain the exception: in the vast majority of systems demonstrated, the final strike command is still issued by the operator.

Over the past year, target auto-tracking technology has progressed from niche experiments to serial solutions being procured by brigades and scaled by the state. But the limitations remain the same: recognition quality depends on imaging conditions, resilience to electronic warfare depends on «last-mile» architecture, and operational effectiveness depends on whether the operator understands when to trust the algorithm and when to take back control.