Recent videos show the much-hyped AI-based “vision” system on Russian Lancet drones resulting in piles of debris instead of the target it was previously confidently aiming for.
Russian Lancet attack drones have an artificial intelligence-based target detection system, but it was disabled due to bugs. Forbes reached this conclusion based on video evidence and expert opinions.
The Russian kamikaze drone “Lancet” is considered one of the best weapons in the Russian-Ukrainian war. Cheap ammunition proved particularly effective in destroying artillery and armored vehicles from a distance. The Lancet was recently updated with a new target tracking mode; This arguably makes it one of the first truly autonomous warfare weapons capable of finding and engaging targets without human assistance.
But there is evidence that the drone did not perform as well as the invaders had hoped, and the autonomous mode was actually turned off. Close examination of videos of the attacks posted on social media shows that the Russian drones were unsuccessful.
Lancets often operate in teams with specialized reconnaissance drones manufactured by ZALA, which also produces Lancet. The reconnaissance team finds the targets and determines their location, transmits the coordinates to the operator who delivers the drones to the field, confirms the target and directs the Lancet to it.
Manufacturers have always boasted about Lancet’s built-in intelligence, and the latest update is expected to deliver a high level of autonomy. In December, military expert Mikhail Kuzovkin told Russian television channel REN TV that target detection is now automatic and does not require operator input. This will make drones invincible.
ZALA CEO Alexander Zakharov does not use the term artificial intelligence, considering development only as automation at the highest level. “There is no artificial intelligence in this drone, only algorithms and decision-making,” Zakharov said in a television interview in July 2023.
An examination of the crashed Lancets revealed that they were equipped with NVIDIA’s Jetson TX2 module, designed to provide “true artificial intelligence” for small, low-power systems such as drones. The credit card-sized TX2 module is optimized for neural networks and machine learning programs with an array of parallel GPUs that process 30 trillion operations per second. The TX2 flies drones like the Skydio 2, a consumer model that’s been called incredibly smart for its ability to chart its own flight path, follow an object, and avoid obstacles without human assistance.
Although the US-made TX2 chip theoretically cannot be exported to Russia due to sanctions, it is available for free on Alibaba in batches of more than 100 for about $200. ZALA probably won’t have much difficulty obtaining these. The only question is how effectively the TX2 can identify objects on the ground without human supervision.
Videos of the Lancet attack are frequently posted on social media and provide insight into the type of targets attacked. About 50% of these are artillery or rocket launchers. The drone’s long range makes it a very effective counter-battery tool.
In several Lancet videos since October, the text “Target Locked On” appears on the screen during the final attack, along with a moving field tracking the target. This phenomenon is also known from modern anti-tank weapons such as the Javelin, where the operator aims at the target before launching. The words “Target captured” appear in three of the five videos about tanks, artillery and trucks. However, this is not the case when it comes to heavier and more concealed targets.
“The Jetson chip appears to be able to support basic object recognition, but I think it’s more a question of how many objects Lancet can recognize and how reliably it can recognize it,” says Zach Kallenborn of CSIS’s Strategic Technologies Program.
However, Samuel Bendett, an expert on Russian unmanned aerial vehicles and a consultant to CNA and CNAS, says Lancet’s claims that they are semi-automatic or fully autonomous are valid. The Russian Ministry of Defense markets these drones as “smart” miracle weapons, but their AI capabilities have now been effectively turned off.
A video dated January 29 shows the Lancet hitting a CV-90 infantry fighting vehicle. The familiar “Target Locked” sign appears and the drone appears to be heading straight for the CV-90 until it veers off at the last second. As one Russian commentator noted, the drone was not targeting an infantry fighting vehicle but a pile of parts.
This is an example of the fragility of AI vision systems that make mistakes that look funny to the observer. Deep learning AI can get confused when an object is at an unusual angle or be fooled by something as simple as subtraction.
There may be many more such losses that few people see because they are not usually published on social media. Interestingly, none of the Lancet videos from the last two weeks feature the title “Target Locked On” or an accompanying bounding box.
The obvious conclusion was that the automatic target recognition software was deployed prematurely and the product was recalled. Previously the inscription appeared not only in complex and intricate shapes such as camouflaged balls, but now it does not appear even on simple targets in the open air.
But the Lancet is getting other changes besides artificial intelligence. Last fall, Lancet gained the ability to remotely detonate without making contact with the target.
Source: Focus
Ashley Fitzgerald is an accomplished journalist in the field of technology. She currently works as a writer at 24 news breaker. With a deep understanding of the latest technology developments, Ashley’s writing provides readers with insightful analysis and unique perspectives on the industry.