How Ukraine integrates machine vision in battlefield drones

BLUF: In 2024, Ukraine introduced machine vision in drones, a function that can recognize objects and hit targets even after losing connection with the operator. Oleksandr Pavlichenko, head of Ukraine’s export control agency, has left his position, he told Counteroffensive.Pro. And a Ukrainian startup Griselda secured investments from Double Tap Investments. And: Arsen Zhumadilov has been officially appointed as director of the Defence Procurement Agency.

THE BIG STORY: Machine vision: New possibilities for target recognition

Ukraine is increasingly using AI-powered drones. In 2024, there was a wave of news about machine vision, which allows drones to recognize objects independently and use this information to hit a target. 

In March 2024, Ukrainian developers successfully tested FPV drones with machine vision in combat conditions, or as they call it, automatic targeting at the end of flight. 

More than ten companies are starting to integrate machine vision into their products. For example, Swarmer is developing swarms of drones that use machine vision to make enemy destruction more efficient (read our story about swarm drone concepts).

What is machine vision?

Machine vision is the ability of a computer (using a drone's camera) not only to see but also to recognize an object. In everyday life, people use machine vision to unlock a phone via Face ID. The phone uses its camera to recognize the face of the owner. 

Machine vision uses a camera that captures an image and converts it into an electrical signal through a sensor. A processor processes this signal and then analyzes it using algorithms that allow objects to be recognized. All these components are connected through a board that ensures their interaction.

“If someone in Ukraine wants to create a genuinely autonomous drone that can navigate independently in space — it is impossible without machine vision. This software can be installed anywhere on the ground equipment, on servers, ships, etc,” Gabe Dublin, co-founder and CEO of Psycraft, a software company developing artificial intelligence for battlefield applications, said. 

Machine vision does not work like human eyes. If a human sees a tree for the first time, they will recognize the object as a tree later, whether leafless, covered in snow, or big or small. 

However, a machine must be trained by showing it hundreds of images of trees to understand that it is a tree. These tree images have to be various: under snow, without snow, at night, during the day, etc., Ilya Ostrovsky, the head of Defense Partnerships in AI Verse, explained. 

The time it takes for the machine to process images depends on many factors, including processor power, which is the most expensive element. 

The practical ‘training’ of machine vision requires two main elements:

• Diverse images (the model must be trained on many pictures of the object to recognize it in different conditions);

• Providing complete image information (each photo must be thoroughly annotated manually, so the AI knows what it shows).

Ostrovsky explained that it takes a lot of time and is expensive — sometimes hundreds of thousands of dollars. Their company has been cooperating with Ukrainian producers since the end of 2024. They create 3D environments where you can simulate how a tank looks on the ground, in water, among trees, and in different conditions — creating hundreds of thousands of variations.

In this way, AI receives many realistic images, allowing it to train models quickly.

“To train tank recognition, we created 100,000 images in different environments - desert, forest, city streets, different lighting, etc. Previously, the AI could not identify a column of tanks. However, after additional training, it can determine their location, distance between them, and angle,” Ostrovsky added. 

How is it used on the battlefield? 

Computer vision is the most effective way to recognize objects where radar does not work, Gabe Dublin said. In particular, for objects on the ground, especially camouflaged ones. 

One main advantage is the ability to operate under Russian electronic warfare (EW). The drones' autonomy in the last few hundred meters to the target is intended to minimize the impact of enemy EW and reduce the chances of operator error.

The state started buying these drones, and hundreds of them went to the army, including products of Vyriy Drone, which won most of Ukraine's tenders for machine-vision drones. 

Still, after feedback from the military, it turned out that these drones were difficult to operate at dusk, so the company supplied more powerful cameras, a representative of Vyriy Drone told Counteroffensive.Pro. 

Detecting large or moving objects, such as tanks, aircraft, warships, or radars, is easier. 

However, it is much harder for machine vision to recognize mines that are small, hidden, and can be shaped differently. 

Ukrainian-Danish startup Dropla, which develops AI-powered demining drones, is dealing with this challenge. The Dropla team teaches their program to notice patterns in images of different types of mines: lines, shapes, etc. 

All pictures are also automatically storyboarded with precise identification of the object and its characteristics. Dropla's model also separates photos by projectile type and uses a specific model for every type of projectile. This helps avoid chaos during training, cause one big model may work improperly with thousands of projectile types. This approach improves recognition accuracy and provides regular updates of mines, one of the technical specialists at Dropla said.

“All machine vision technology is based on the camera and its image quality. If there is any communication interference, or you choose the wrong angle relative to the target, the drone will produce an error,” Vladyslav Oleksienko, co-founder of the drone manufacturer Smart Electronics Group, added.

What are the main obstacles to application?  

There is a big problem with training AI to recognize objects in the defense sector due to a lack of the most actual data. In the Russian-Ukrainian war, new types of drones, camouflage, and vehicle models constantly appear. AI needs many pictures to be able to recognize these new products. 

Ukrainian Delta situational awareness system developers are also trying to speed up the development of AI at the front. The Ukrainian Ministry of Defense said that for years, the military has uploaded photos and videos of Russian equipment from the frontline to Delta with the help of the Avengers system. They plan to use this dataset to help Ukrainian developers train their neural networks. 

To further develop the system, it must be used on the frontline on a massive scale. Vyriy Drone received a government contract in the summer and delivered its drones in early winter. So they are only now starting to be handed over to the military. 

Like any other device, this system can be subject to hacking, but it is designed to minimize such risks. For example, when a drone operates autonomously, it cannot be intercepted because it lacks external control, Gabe Dublin said.

Another problem is the risk of friendly fire. Artificial intelligence does not think like a human, it analyses patterns, making it challenging to distinguish enemy equipment from your own. For example, if Ukraine did not use Soviet equipment, it would be much easier to distinguish Ukrainian tanks from Russian ones. This means that planning and scouting remain the most important thing, Serhiy Kuprienko, founder of the US-based company Swarmer, which develops swarm technology with machine vision, explained.

One possible way to avoid friendly fire is to integrate a radio signal recognition system when a drone "asks" the vehicle: ‘Are you ours?’ and receives a response, Dublin suggested. Still, for the technology to adapt to the needs on the battlefield, it must be constantly used, Ihor Raikov, a platoon commander in an anti-tank company of the 13th Brigade of the National Guard's Khartiia, noted.

In November 2024, the government procured 3,000 FPV drones with machine vision and targeting technologies. Reports also suggested that the procurement would be expanded to 10,000 units. 

“This technology has prospects. A drone with machine vision is only 1,000 UAH ($20) more expensive than a regular one. Our military is very resourceful, so they will find a use for it, even if it is not perfect,” Oleksandr, a representative of Vyriy Drone, said. 

BUSINESS LUNCH TOPICS: What our reporting shows!

1. State Service Export Control left without a head

Oleksandr Pavlichenko resigned as the head of the State Service for Export Control, which provides permits and licenses for exporting or importing military or dual-use goods from and to Ukraine. 

His appointment came with the end of a five-year contract — the Cabinet of Ministers did not offer to extend Pavlichenko’s tenure, he told Counteroffensive.Pro.

“We had various conversations about this [the contract]. It was decided that at this stage it was enough. It did not come as a surprise to me,” Pavlichenko explained.

Before the outbreak of full-scale war, the SSEC chairman was selected through an open competition. However, this has been suspended for the duration of martial law. The Cabinet of Ministers will appoint the new head on the proposal of Yulia Svyrydenko, First Deputy Prime Minister and Minister of Economy.

Pavlichenko is open to new kinds of work that may be useful for those seeking expertise on export controls. 

“I will continue to pay attention to the changes [in the SSEC] and share my experience. Many of the things we have developed may be useful in the future”, he said.

On Saturday, we will publish an interview with Pavlichenko about his thoughts on export control policy and the future of SSEC.

2. Arsen Zhumadilov officially became director of the Defence Procurement Agency

On March 6, Minister of Defense Rustem Umerov officially appointed Arsen Zhumadilov as Director of the Defence Procurement Agency (DPA), ending the saga of Maryna Bezrukova's dismissal (read about the origins of this conflict at this link). 

DPA purchases and supplies lethal weapons to the Ukrainian army. The main task for Zhumadilov at the DPA will be to make the procurement of weapons, ammunition, and equipment as efficient, transparent, and quick as possible, said Minister of Defense Rustem Umerov.

Zhumadilov had been acting director of DPA since Bezrukova's dismissal in late January, along with state representatives on the Supervisory Board, Taras Chmut and Yuriy Dzhygyr.

Along with his appointment to the DPA, Zhumadilov was relieved of his position as Director of the State Logistics Operator (SLO), a non-lethal supplier to the Armed Forces. The decision to dismiss Zhumadilov was supported by the SLO Supervisory Board on March 6.  Now the SLO is temporarily headed by Andrii Sozanskyi, SLO's executive director.

3. Finnish Liquido VC to allocate “millions of euros” to European defense startups

Finland-based venture fund Liquido plans more investments in European security, The fund intends to allocate “millions of euros” to European defense startups in pre-seed and seed stages, Jussi Muurikainen, founder of Liquido venture fund told Counteroffensive.Pro. 

The company will particularly look for developers who have already proved their products on the market, and have business working models or clients. As for startups founded in Ukraine, Liquido does not demand registration abroad.

“We don’t want the brightest minds of Ukraine to leave the country. We want them to come back to Ukraine to build the future in Ukraine,” Muurikainen said.

4. Double Tap Investments funded Ukrainian startup Griselda 

Ukraine-based startup Griselda secured a strategic investment of $600,000 from Double Tab Investments, a Finland-Ukraine VC. 

Griselda develops a solution to transform battlefield intelligence processing data faster and with precision. Among the advantages of Griselda are the fast adaptation of the platform, which allows providing a separate product for each client, emphasized Oles Khudoba, co-founder of Double Tap Investments. According to him, many Ukrainian units already use the Griselda solution. So the VC had no doubts about investing in it. 

“Griselda is a more understandable and accessible analog of Palantir [American developer of AI-based software for analytics] in many aspects, and we want Ukraine to have its own. We have been working with them for over a year,” told Khudoba. 

ON THE MOVE: Professional movement, promotions and industry news.

Know someone in the defense tech space who has made a professional move? Drop us a line at [email protected]! 

• Lilia Slobodian became an international relations manager in Tech Force in UA, a coalition of over 50 private manufacturers. Previously, she worked as an international aid analyst at the Embassy of Canada.

• Farsight Vision, an autonomous system's developer, is hiring a business development specialist.

30 SECOND ELEVATOR PITCH:

Dwarf Engineering is a Ukrainian manufacturer specializing in the full-cycle development and production of autonomous systems for the frontline. Their flagship product is Narsil, an autonomous drone control system that minimizes pilot workload and transforms an FPV drone into a ‘satellite.’ Narsil enhances UAV resilience against electronic warfare. It is compatible with FPV drones ranging from 7 to 25 inches, with future integration planned for fixed-wing aircraft, munitions, and robotic systems.

“We are seeking international partners to export Ukrainian expertise to other democratic countries. The company is open to military and civilian contracts abroad, starting from $500,000, and offers autonomous control systems", said Vladyslav Piotrovskyi, CEO of Dwarf Engineering.

MUST READS:

• Colt CZ, a firearm producer, is now producing up to 400 Bren 2 assault rifles per day at Ukrainian plants. They are aiming to replace AK rifles in the Ukrainian army.

• Ukraine signed a memorandum of understanding with Diehl Defence, the producer of the IRIS-T air defense system. The two parties plan to increase the supply of anti-aircraft missiles and air defense systems to Ukraine three times.

• Ukraine’s government is planning to purchase 4.5 million FPV drones in 2025. It’s three times more than last year. 

• Russia has updated the T-72 tank, including Relikt reactive armor, which consists of blocks of explosive materials. Explosives reduce the possible damage to the shaped charge by detonating the inner explosive. It won’t cause any additional damage, because the tank has thick armor.

• The Netherlands will invest in Ukrainian drone production. Amsterdam agreed on an investment package worth 700 million euros for drones, part of which will be produced in Ukraine.

• The Norwegian parliament plans to increase military aid to Ukraine from 3 billion euros to 7.5 billion euros.

• Ukrainian manufacturer of FPV components 603700 sent the first batch of flight controllers for FPV drones to the U.S. Currently they are being tested and soon will be sold through local distributors.

Thanks to our media partner: