Unmanned systems instead of soldiers: Ukraine bets on AI-supported autonomous warfare

In Ukraine, unmanned systems are increasingly meant to take soldiers out of direct combat and replace them there autonomously. The Ukrainian armed forces are thus pushing ahead with the transition towards more automated warfare.

This is about more than protecting scarce human resources. Modern battlefields make growing demands: soldiers become fatigued, suffer psychological stress and are overwhelmed by ever larger volumes of data from a wide variety of sensors and sources. The use of artificial intelligence (AI) is intended to solve these problems.

Kateryna Bondar has analysed in depth in a 37-page report what an AI-supported military strategy could look like in future. She works at the Wadhwani AI Center of the Center for Strategic and International Studies (CSIS) in Washington, D.C.

The report quotes a Russian soldier: “It is a drone war”. He first saw a Ukrainian fighter only after he had already surrendered to an unmanned aerial vehicle (UAV).

What defines an autonomous weapon system

The US Department of Defense defines an autonomous weapon system as a system that, once activated, can select and engage targets on its own – without further human intervention.

These must be distinguished from non-autonomous or semi-autonomous systems. Although they automate individual steps such as detecting or tracking a target, they do not carry out the entire decision-making and engagement process on their own.

Drone production in figures

According to Defence Minister Rustem Umerov, Ukraine produced around two million drones in 2024, more than 1.5 million of them first-person view (FPV) drones. Domestic production accounted for 96.2 per cent of the drones used by the Ukrainian armed forces. Over the same period, the country thus sourced around 80,000 drones from abroad.

Between January and September 2024, the Ukrainian military approved 140 UAV models and 33 models of unmanned ground vehicles (UGV) for operational use. For 2025, Ukraine plans to produce around five million drones.

AI evaluates reconnaissance data

One field of application for AI is the evaluation of data from intelligence, surveillance and reconnaissance (ISR). Algorithms prioritise images and videos according to how many events are observed and how intense they are – such as ongoing fighting or larger movements of troops and equipment. In this way, objects are to be detected and classified. The imagery comes from satellites, drones and fixed cameras.

AI is also applied to intercepted communications such as voice and text messages. Thousands of group chats and messages run daily via messengers such as Signal and Telegram, frequently containing unstructured, non-standardised reports from the battlefield. In addition, there is intercepted audio communication from various sources, which can also contain valuable information.

Transcribing these volumes of data in time and evaluating them so that usable insights emerge is hardly possible without technical help.

Griselda for voice and text data

For this unstructured data, Ukraine relies on the AI-controlled system Griselda. The platform is specifically designed to analyse large amounts of voice and text information.

Zvook listens for threats from the air

Different sound patterns can be classified and identified with acoustic systems. The acoustic detection system Zvook has significantly improved Ukrainian air defence. High-quality microphones provide the sound analysis used to detect aerial threats at low altitudes – precisely where radar systems lose effectiveness.

Zvook's acoustic sensors are arranged in a grid and currently monitor around 20,000 square kilometres of Ukrainian territory.

Background noise such as cars, birds or human voices is filtered out by AI. Based on the typical acoustic signature, it classifies aerial threats such as aircraft, helicopters, cruise missiles or piston-engined drones.

Thanks to Zvook, Ukrainian radar installations can remain in standby mode; they are only switched on automatically when there is a specific threat.

Automatic target acquisition by AI

At the heart of AI-supported automatic target acquisition is the ability to detect and classify targets faster and more accurately than human operators. Trained on extensive datasets from combat scenarios, advanced machine learning algorithms distinguish between vehicles, human figures and structures – across highly varied terrain and under changing conditions.

In combat, automatic target acquisition currently reaches an average of around one kilometre; under optimal conditions, up to two kilometres are possible.

For the AI to adapt to the dynamic conditions on the battlefield, it must be continuously retrained with real combat data.

Unmanned systems under electromagnetic interference

When conventional control systems are disrupted by electromagnetic warfare, the success rate of target acquisition drops to around 10 to 20 per cent. The main factor here is interference with signals from global navigation satellite systems (GNSS) such as GPS, as well as interrupted control links between the drone and its human operator.

Autonomously navigating drones, by contrast, achieve a success rate of 70 to 80 per cent.

A distinction must be made between solutions for navigation over the “last mile” and systems for long-range missions intended to penetrate deep into enemy territory.

13 березня дрони СБУ атакували нафтопереробні заводи в РФ. відео з соцмереж pic.twitter.com/8HG8MjS6Ak

For long-range reconnaissance missions, AI-controlled navigation without GPS has now become a key element. Flight routes with more than 1,000 waypoints are programmed so that the drones bypass enemy air defence systems.

The drones determine their position by autonomous navigation using preloaded satellite imagery and terrain data. This allows them to approach targets with high precision even without active satellite communication.