AI‑driven FPV drones shift balance in Ukrainian conflict
The war in Ukraine is the first conflict where drones are used on such a massive scale by both sides. Initially, Ukrainian drones were tremendously effective, but their efficiency drastically dropped due to the widespread use of jammers by the Russians. However, for several months now, supremacy seems to be returning. This is due to the implementation of artificial intelligence in FPV drones.
Ukrainians are currently using hundreds of thousands of drones for various tasks, but the most notable recordings are of FPV drone attacks. These drones are used to target equipment like BMP-3 infantry fighting vehicles, tanks, including the latest T-90M, and even individual soldiers or helicopters.
After the initial tremendous effectiveness of Ukrainian drones, Russians began to deploy jammers massively, affecting control signals or GPS signals for drones. This created a classic shield and sword race, with alternating periods of drone dominance and impotence. During this time, experiments with control signal frequencies were conducted.
A solution to these problems was sought, as seen with professional military drones, in artificial intelligence algorithms enabling autonomous attacks on designated objects. After months of trials, Ukrainians, sometimes with Western companies' help, achieved satisfactory results. Below, you can see a crashed FPV drone supported by a Google Coral AI module.
FPV drones with artificial intelligence — only physical elimination ensures protection
The key is providing a module with enough computing power to create computer vision that tracks and follows a selected object. Due to the limitations of vehicle-mounted jammers, their effective range is a maximum of 400-500 metres, and fully autonomous flight must be sufficient to cover such a distance.
Today, many commercial AI solutions are quite capable of object recognition. The main difference is that instead of recognizing humans or, for example, cute kittens, the algorithm must recognize a T-72 family tank and follow it. Some solutions, like Skynode-S modules, even provide navigation based on triangulating the drone's position by comparing the terrain seen by the drone’s camera with preloaded satellite maps.
One of the challenges is developing machine learning algorithms, and another is ensuring systems have sufficient performance, compact dimensions, and low power consumption. As a result, a drone is created that can only be stopped by shooting it down.
In the case of professional solutions, these include Israeli SMASH modules from Smart Shooter, which have gained significant interest in recent years within the armed forces and enable a drone to be shot down with handheld firearms even from a few hundred metres away (effective range varies with the calibre of the weapon it's mounted on).
Meanwhile, in the case of improvised solutions, 12-gauge smoothbore shotguns perform well. These, coupled with appropriate buckshot ammunition, can be effective even at distances of around 50 metres. With the use of new ALDA ammunition from Beretta Holding, it is possible to hunt drones at distances of about 80 to 120 metres.
Google Coral AI development board — the perfect solution for prototyping
The photos show that part of the drone includes a G950-04742-01 development board characterized by a performance of 4 trillion floating-point operations per second (TOPS) while requiring only 2W. For example, Google boasts that this system can execute the MobileNet v2 SSD neural network architecture instruction in 14 ms, whereas a 3 GHz Intel Xeon Gold 6154 processor needs 106 ms.
The Google board features a ready-made NXP's iMX 8M system with four ARM Cortex-A53 cores, an ARM Cortex-M4 core, integrated Vivante GC7000Lite graphics, Google’s proprietary Edge TPU coprocessor, and a cryptographic coprocessor. Additionally, it has 8 GB of eMMC memory, 1 GB of LPDDR4 memory, and necessary I/O ports.
These boards are available for purchase freely, and Ukrainians, unlike Russians, have unhindered access to them. However, as the sanctions reality shows, a certain number will likely also be available in Russia.
