Russian Kamikaze Drone Now Seems To Identify Its Own Targets

The yellow text at the top of the screen identifies the target of this Lancet kamikaze drone as a … [+] Leopard 2 tank. This looks like automated object recognition.

Russian MoD

We previously described how Russia’s Lancet loitering munition had been seen with a ‘target lock’ mode. Depending on who you believe, this either allowed the operator to lock on to a target once they had seen it, or enabled the drone to autonomously identify and attack targets on its own. The 25-pound Lancet can destroy tanks, artillery and other vehicles from 30 miles away and the new mode looked like a real upgrade – but it disappeared on later videos. An incident in which the lock shifted away from a vehicle to target a pile of debris may explain why.

As expected, ZALA who make the Lancet have been tinkering. The target lock appears to be back, and in smarter form, possibly foreshadowing a new level of AI-enabled warfare.

Target Identified

The Lancet here locks on to one T-72 tank in a row of several

Russian MoD

A slew of videos recently posted on the LostArmour site, which collects Lancet strike videos from various miliary Telegram channels show what looks like a new version. Instead of the text ‘target lock’ in Russian, the name of the vehicle appears on the display — though the clips supplied are very short screen grabs were necessary to pick out some of them. This strongly suggests some form of automated object identification is at work.

“ML [Machine Learning] classification seems the most likely explanation,” Zak Kallenborn of CSIS’s Strategic Technology Program told Forbes. “The classification being displayed is curious though. Is that for the benefit of the operator, perhaps to verify the accuracy?”

The new display also shows a red +-sign when the target is identified, plus other new symbology. In this video the target is identified as a 2S1 Gvozdika, a 122mm self-propelled gun. In this one the target is a 2A65 ‘Msta-B’ 152mm self-propelled howitzer according to LostArmour, this strike is on ‘unidentified equipment’ – but the Lancet video flashes up text identifying it as a US-supplied M109 Paladin. Here one identifies and hits a Leopard 2 tank, this one selects one of several T-72 tanks in a row.

Here the Lancet locks on to a U.S.-supplied AN/TPQ-50 radar vehicle

Russian MoD

The Lancet strikes shown still tend to follow the usual format: we see the target from the viewpoint of a reconnaissance drone, then we switch to the view from the oncoming Lancet as it approaches the target – the video quality is noticeable worse – then then back to the recon drone to see the effects of the strike. So the Lancet is not working on its own.

All the videos which include the new target identification come from a single batch of 21 Lancet strikes, posted on the same day. These all came from the same source.

Note the difference in appearance with the old-style ‘Target Lock’ Lancet display

Russian MoD

Special Operations

The clips of Lancet strikes were extracted from a nine-minute video posted on the Rybar Telegram channel on February 27th. This is claimed to show the work of Russian special forces units operating behind Ukrainian lines — “not only in the Chernihiv and Sumy regions, but also along the entire line of contact, there are even strikes in the Odesa region.”

The very brief clips of Lancet strikes are shown among a lot of other footage of Russian forces fighting in woodland and ambushing Ukrainian soldiers in cars and pickup trucks, including graphic images of dead bodies. One Lancet launch is shown. Interestingly, there are no smaller FPVs visible – these troops appear to be equipped entirely with high-end Lancets (costing around $35,000 each) rather than the smaller option.

These Russian special forces to rely on the Lancets for attacks on armored vehicle, as they do not appear to use heavy anti-tank weapons, and the Lancet strikes cause far more damage than anything else. This makes sense for a unit operating behind enemy lines as the lack of firing signature (no blast, flash or smoke) means that the source of a Lancet cannot be traced, and they can be launched from miles away.

The new target identification has two potential benefits: it may work faster and more reliably than a human observer in confirming targets, and it will continue to work even if the link with the operator is lost due to jamming.

Unsurprisingly, it only appears to work with military vehicles. There are no Lancet strikes on trucks, building or personnel for comparison. And there is no identification shown when it his a Vampire vehicle, possibly because these are too new and rare to be in the database.

New Paradigm

We cannot necessarily take anything from a Russian source at face value, and the Rybar Telegram channel reportedly has close links with Russia’s Ministry of Defence and the FSB intelligence service. The 21 hits may not be representative of all the launches. We do not know if this is a 90% hit rate of 10%, or how many Lancets veered off course to target boulders or trees.

On the face of it though Russia is fielding a drone that can identify and lock on to specific types of vehicle.

“If this is actually the case, then we are witnessing the capability that the Russian military has announced a while back, including stating that these drones will take to Ukrainian battlefield,” Samuel Bendett, an expert Russian drones and adviser to both the CNA and CNAS, told Forbes.

However, the smart Lancets are not simply being flown into an area to find targets. The pattern is still very much that target was previously spotted by a scout drone operator and the Lancet is vectored on to it. The lock is used for terminal guidance rather than seek-and-destroy, though Bendett says it is impossible to judge exactly how it is working.

“But judging from multiple recent MoD statements about AI-enabled drones, this capacity is in the field at least to a certain extent,” says Bendett.

Russian soldiers preparing Lancet for launch

Russian MoD

The AI Arms Race

This capability is not unique to Russia. In recent years AeroVironmentAVAV
, makers of the Switchblade series of loitering munitions, have revealed that they have developed an autonomous version .

In January 2023, Brett Hush, vice president of tactical mission systems at AeroVironment, told DefenseOne that “We’ve demonstrated with the DOD our ability to do that to identify like… 32 tanks,” but noted that autonomous strikes without a human in the loop would not be allowed under current Pentagon rules.

Last year Wahid Nawabi, AeroVironment’s CEO, told AP news that “The technology to achieve a fully autonomous mission with Switchblade pretty much exists today,” but that he expected it to be three years before policy changed to allow weapons to find targets without human involvement.

Ukrainian drone developers are working to field similar capabilities; the Saker drone with autonomous attack capability has already been used on a small scale.

A report from UK defence thinktank RUSI this week suggests that victory in Ukraine will go to the side which succeeds in fielding large number of autonomous drones, and that Western efforts should go to supporting Ukrainian efforts in this direction:

Critics should be reminded that if Russia masters the employment of thousands of cheap ‘slaughterbot’ drones that use AI for targeting on the battlefield, as it is now attempting to do, the Ukrainians will pay the high price of losing the war and ivory-tower moral hubris in Western capitals and universities will be complicit in this strategic failure,” states the RUSI report.

An AI arms race is under way in Ukraine. Its impact will be felt far beyond the borders of Europe.

Read More