Russian AI Drone Swarm? Not There Yet – A Guide to Drone Navigation

Russian drones used artificial intelligence (AI) to defeat electronic warfare (EW) defenses, recent reports claimed.

“[Russia is] even using AI to make it more difficult to jam their Shahed drones,” Kyiv’s EU Ambassador Vsevolod Chentsov told Politico in late December.

“They’ve started equipping them with the capability to recognize targets even if they’re disconnected from the network, so if they’re flying offline they can still see the object that resembles the power station,” he said.

A Wednesday opinion piece for Breaking Defense caught on Chentsov’s comments. However, citing an unnamed Ukrainian military official, it suggested that the new capabilities are not AI but “more conventional than it appears,” though the author still raises concerns about advancing Russian drone capabilities.

The opinion argues that “when autonomous navigation merges with automatic target recognition,” a drone can engage targets without human inputs, ultimately paving the way for AI drone swarms.

Good news for Ukraine: Russia is unable to do most of that just yet – but to understand where Moscow is at with their technology, one should understand what drone swarms are, how drones navigate, and how they recognize the targets.

What is a drone swarm?

Other Topics of Interest

Ukraine Dismantles Nearly 50 Smuggling Networks Aiding Draft Dodgers

Kyiv has clamped down on draft evasion in recent months.

Drone swarms refer to multiple drones that can fly cohesively to achieve their mission.

“Drone swarms integrate advanced computer algorithms with local sensing and communication technologies to synchronize multiple drones to achieve a goal,” read a 2023 report by the US  Government Accountability Office (GAO).

By definition, AI drone swarms mean the drones would be able to communicate with each other, understand their coordinates, and react to changes autonomously through AI.

The Breaking Defense article said the discovery of a modem mesh module – which enables decentralized wireless communication – on Russian Gerbara drones showed “early signs” of onboard communication between drones that enables the swarm to “reroute and remain effective.”

That said, there are still hurdles in navigation and target recognition that prevent Russia from achieving AI drone swarms.

How do drones navigate?

Target coordinates – and likely flight paths – are normally preloaded onto the drone before launch.

To go from one point to another, a drone must understand where it is and where it has to go at all times, which is aided by various guidance technologies – inertial navigation system (INS), satellite triangulation and other methods to determine its present location.

Imagine flying from New York City to Los Angeles – going west 4,000 km (2,485 miles) sounds easy enough, but with the mountains in the way, you need to turn eventually – which requires reference waypoints.

INS

You can memorize all the turns and steps needed, assuming both the starting point and destination are clear – but you still need to know where you are.

The INS tells you where you are based on how you moved from the initial waypoint. It relies on acceleration sensors, rotation sensors (gyroscopes), and other equipment to calculate your movements in three axes (fore-aft, left-right, and up-down), to tell it how to get to each waypoint.

If you can’t follow along with a map because the clouds, or smoke from wildfires, are below your aircraft, you might use an INS. The INS helps account for small variations in flight parameters, effects of wind, and other real-world conditions, including the earth’s rotation by sensing movements in all three axes.

Unfortunately, without updates by external confirmation of its position, an INS will incur errors that accumulate over time, and thus it would drift off course by default

One wrong turn or simply not being on a very precise course can throw off the remaining calculations, with you ending up in Long Beach or Malibu, not at Los Angeles International.

For Shaheds, one of the US-made Micro-Electro-Mechanical Systems (MEMS) accelerometers found on the drones are known to be susceptible to temperature fluctuations and interference that could lead to inaccurate readings, thus bringing them off course.

Satellite triangulation, or Global Positioning System (GPS)

You can also take out your smartphone and find your coordinates using Google Maps and where to turn – that would be satellite tracking, the most common being Global Positioning System (GPS).

Some countries have their own version of GPS, such as Russia’s GLONASS, the EU’s Galileo and China’s BeiDou systems.

In short, GPS functions by measuring the time it takes for signals from multiple satellites in space to arrive at your position. This determines your location through triangulation after calculating your distance from each satellite, whose position your GPS receiver, i.e., your phone, already knows.

Your phone’s GPS receiver tells you where you are, but the app tells you when and where to turn – GPS merely provides a reference point, not guidance.

As such, GPS provides drones with external references, telling them where they are to help them adjust the flight path accordingly.

But drones relying on GPS also have an inherent flaw: The signal can be drowned out (jamming) or disrupted by counterfeit signals (spoofing), leading the drones to believe it is in the wrong location and throwing it off course – this is commonly done in Ukraine against Russian drones.

Visual or radar confirmation

What about the old-fashioned way of following road signs and landmarks, making the turns accordingly?

The Breaking Defense article suggested that Russia upgraded its drones by using a “new generation of Digital Scene Matching Area Correlator (DSMAC)” based on Chentsov’s description.

In simple terms, DSMAC cross-references preloaded terrain images against what it sees using a camera while flying to determine its location.

A research paper on the US Tomahawk missile, which uses a combination of GPS, DSMAC and Terrain Contour Matching System (TERCOM) to measure positions en route, said DSMAC is generally the most accurate among the three.

Breaking Defense speculated that Russia uses “computer vision algorithms rather than advanced AI.”

However, DSMAC requires a clear line of sight between the drone and the ground to function, meaning adverse weather conditions can affect its performance.

TERCOM, normally found on cruise missiles, uses onboard ground-mapping radar to match the terrain beneath and help negate the issue. There is no evidence that Moscow is using that technology on its drones.

Do drones recognize targets?

Shahed drones, like missiles, home in onto the coordinates, not objects – but the Breaking Defense opinion suggests the onboard computers are “loaded with an AI model trained to recognize energy infrastructure.”

Hypothetically, the Shaheds may be equipped with cameras and onboard processing chips to help identify the targets – power stations, for instance – that help improve accuracy during the endgame stretch of their flight.

As the opinion piece pointed out, the idea is “not new to modern striking drones.” Similar features are used in Ukrainian first-person view (FPV) drones to help them fly at the target autonomously when the pilot loses control due to EW jamming.

But that is also where Chentsov’s comments became questionable – he specifically referred to the long-range drones (such as Shaheds) used against Ukrainian energy infrastructure, which, unlike FPV drones, fly on a pre-planned route and are not controlled remotely.

But assuming Chentsov is correct in saying the drones can recognize the targeted objects autonomously, the Breaking Defense opinion piece said it represents a “threat upgrade” nonetheless, where the onboard software can theoretically be trained to recognize any protective structure Ukraine may have erected and identify weak structural points.

Autonomous navigation, targeting?

The key aspect of true autonomy is the ability to interpret the surroundings and react accordingly – which the Russian drones cannot achieve at present.

For drones, autonomous navigation means they would need to be able to calculate the optimal route based on the environment – adjusting their flight path on a pre-planned route isn’t exactly autonomous navigation.

It would also involve reacting to surface-to-air and air-to-air threats along the route as well as atmospheric conditions, including weather, fire and smoke, such as the wildfires currently blazing in Los Angeles.

Additionally, it would have to deal with conditions generated by the explosion of other drones in the swarm and the damage inflicted by them in the target area.

While Russia appears close to achieving autonomous targeting based on Chentsov’s comments, it still requires getting to the destination in the first place – a drone might know what a power station looks like and where it is, but it still has to see the actual facility before hitting it.

That said, the Breaking Defense opinion noted Russia’s reliance on Western components and Moscow’s pursuit of AI advances through cooperation with its allies, which highlights the need to bolster sanction efforts.

Technology is a combination of hardware and software – if Russia continues to get its hands on Western hardware while simultaneously finding ways to advance its AI capabilities, AI drone swarms may not be that far off.