Destroying Russian Tanks Is Just The Start For U.S. AI Drone Autopilot

The FPV video shows the view from a Ukrainian kamikaze drone closing in on a Russian armored vehicle. The video feeds cuts out, broken by Russian radio jamming and the pilot loses control. That would normally be mission over, but video from a second drone shows the AI-enabled FPV guiding itself straight to the target and scoring a direct hit. It is one of the first strikes using a new smart autopilot supplied by U.S. company Auterion which shrugs off jamming and has a higher hit rate than human pilots.

The Skynode S autopilot was unveiled last week but is already on the frontline in Ukraine. The ability to lock on from long range makes jamming useless; I have seen a number of combat videos which I cannot share for security reasons, and caveats apply, but as far as can be judged, the system looks effective. As Auterion CEO Dr Lorenz Meier told me though, hitting targets automatically with FPVs is just the start.

Auterion’s app store allows developers to write their own apps for Skynode S as well as downloading what they need. Meier says terminal guidance apps for attack drones are currently the most popular.

“Terminal guidance is like MS Office on a laptop,” says Meier. “It is the minimum requirement which everyone expects.”

Open-Source Drone Intelligence

Auterion has a long pedigree in the drone world, having led an open-source movement for more than a decade. In 2008 Meier, then at ETH Zurich, set out to build an autonomous drone. The result was Pixhawk 4, universally known as just PX4, an open-source flight controller giving drone developers a flexible set of tools. There are now 10,000 PX4 developers worldwide, including some who use it for robots and other machines.

The Pentagon clearly believe in Auterion’s open-source software as they put it at the heart of their Blue sUAS initiative for U.S.-made small military drones. This was implemented as alternative to Chinese-supplied drones which come with a built-in security risk as standard and are banned by the U.S. military.

Auterion previously developed the Skynode X, a powerful autopilot and mission computer to give drones autonomy with including machine vision, object recognition, visual navigation and a variety of third-party apps. The newly-released Skynode S offers similar capabilities but at greatly reduced cost.

“We are offering a ‘Ukraine Aid’ price point in Ukraine and have licensed the software free of charge to the government of Ukraine,” says Meirt. “Generally it is priced in the range of an Android phone, mid-hundreds of U.S. dollars.”

This puts Skynode S in the same class are thermal imaging in terms of adding significantly to the cost of a $500 FPV, but giving greatly enhanced capability. And the cost may go down as production is scaled up.

“Skynode S … will be produced in the tens of thousands, introducing unprecedented scale,” says Meier.

It is no surprise to learn that Auterion are also involved with the Pentagon’s Replicator program to field large numbers of low cost AI-enabled drones.

100% Lethal (So Far)

The guidance system provides optical lock-on: the operator identifies the target and flags it for the autopilot while the drone is well outside jamming range. Then it can carry on through the ‘jamming bubble’ even if the operator loses contact. According to French Army Chief of Staff Gen. Pierre Schill, 75% of drones are currently brought down by jamming.

Meier says that their AI terminal guidance system used in Ukraine has scored a 100% hit rate so far, compared to 20%-40% for FPVs controlled by a human. He does not expect to maintain an unbroken run of successes but believes they can achieve 90% in the longer run.

This means the two biggest causes of a miss—pilot error and jamming—can be eliminated.

The implication is that with a guidance system like Auterion S, four times as many drones could get through, and would then have a higher hit rate thanks to smart guidance. It is a sobering thought that the already high FPV kill rate might be boosted so dramatically by such a simple modification.

And while the current version simply guides the drones to the target, future developments are likely to feature aim-point selection, where the AI picks the most vulnerable spot. For example they might circling round to hit the turret rear on Russian tanks which tends to cause a catastrophic explosion, or the ammunition storage on a self-propelled gun.

The open-source software makes it easy to develop new applications. In June Auterion’s Dronecode nonprofit hosted a 48-hour hackathon in Krakow for NATO, challenging teams of coders to develop a system to visually identify targets and plot a flight path to intercept them. They used an early version of Skynode S with AuterionOS, similar to the product just released.

“This was a world first,” says Meier. “The developers were able to focus simply on developing the app.”

But the technology is capable of far more.

Boosting Bombers And Aiding Interceptors

GPS and other satellite navigation is heavily jammed in both Ukraine and Russia, affecting even some military-grade systems and throwing U.S.-supplied Excalibur guided artillery rounds and JDAM guided bombs off course as well as drones. The usual solution to this is expensive inertial navigation modules, but these become inaccurate drift over longer ranges and are expensive.

AI systems are also capable of visual navigation, using a drone’s camera to find the way by matching objects on the ground, much like early pilots, using images from previous reconnaissance flights. Some Ukrainian drones already use AI software called Eagle Eyes to do this.

Meier says Skynode S can add an extra twist: terminal guidance. Once the drone has reached the target area, an object recognition app can visually acquire the target and hit it with pinpoint accuracy. In principle this means even a strike from several hundred miles away can find a specific piece of machinery at an oil refinery, strike the center of a radar dish or go through a particular window. Or simply homing in on a tank parked far behind the frontline which has been spotted via satellite.

This setup has already been tested on Ukrainian drones equivalent to the Russian Lancet and Shahed.

“The testing proved that it can accurately hit targets,” says Meier. “The system will field to the frontline in the next weeks.”

Skynode S can also help with air-to-air combat. Meier expects that autonomous ‘dogfight’ apps to be available shortly which will enable an FPV drone to plot an intercept course, outmaneuver an opponent attempting to evade and destroy it without any operator input, detonating when it gets to within kill distance.

Ukrainian FPV drones recently started intercepting Russian reconnaissance drones at high altitude. Skynode S might significantly the increase number and success rate of these engagements, while also ensuring that Ukrainian drones can outfly Russian interceptors.

An AI pilot might help with other tasks too. For example, some skilled FPV pilots are able to carry out dive bombing attacks, giving the precision of a kamikaze strike but without sacrificing a drone. The main limit seems to be the lack of experienced pilots. But a dive-bombing app could soon be just a download away.

Automated drone minelaying, or using jam-proof drones to drop tire-shredding caltrops on specific roads behind enemy lines, or just resupply friendly forces could also get much easier.

The Bigger Picture

Looking at the bigger picture, Meier says Skynode S is already integrated with Ukraine’s battlefield management software such as Delta and Kropyva. These command-and-control systems collate information from drones, satellites and other sources and combine it into a single picture visible to commanders on the ground. He demonstrates this with a video of an AI-powered reconnaissance drone spotting objects and feeding data back.

Targets are always verified by a human operator, but the AI system has already done the vast majority of the work in finding items of interest from the video feed. Because each drone no longer needs an operator, this will provide more detailed 24/7 surveillance of the battlefront and in real time. And because it only needs to communicate when it finds something, AI-enabled reconnaissance requires far less bandwidth to send far more useful information.

AI also offers the possibility of incorporating capabilities like change detection, automatically comparing a scene with how it looked last time around to spot new minefields, trenches, vehicle tracks or camouflaged positions.

Putting AI at the edge could dramatically accelerate to the process of gathering, processing and disseminating battlefield intelligence. Thousands of smart, autonomous reconnaissance drones may ultimately be more significant to warfighting than terminal guidance.

The Name Is Skynode, Not Skynet

Skynode S also supports drone swarm control: a single operator can control multiple drones. So one operator might have a several reconnaissance drones and a stack of FPVs ready to launch, using the scouts to find targets and vectoring in the FPVs to engage them.

This makes it sound as though the whole kill chain could be automated. It is important here that there is always a human on the loop supervising operations, even if they no longer need to be in the loop.

Meier stresses that Skynode S is not about creating autonomous killing machines. He compares the level of automation to the way that guided missiles are used at present. The FPV operator locking on to a target is doing exactly the same thing as a Javelin missile operator: both are ‘fire and forget’ weapons which home in on targets previously designated by an human.

“I would have serious concerns about anything more autonomous,” says Meier. “But there is still a lot more than can be done without allowing drones to autonomously pick targets.”

It will take some work to explore fully the new capabilities offered by Skynode S. But with a large pool of software developers, low-cost hardware, and a ready supply of drones to fit it to — and a war to win — things are likely to happen fast.