Autonomous drone defeats human champions in racing first
tudelft.nl
tudelft.nl
A few details I picked up:
* The drones are a spec drone across the league. It's a fairly large-footprint FPV racing drone (it's a 5" propped drone, but it's very stretched out and quite heavy) with both a Betaflight flight controller and a Jetson Orin NX onboard. Teams were only allowed an IMU and a single forward camera.
* It's unclear to me whether the teams were allowed to bypass the typical Betaflight flight controller which is present on the drone and use direct IMU input and ESC commands from the Jetson, or whether they were sending and receiving commands from the flight controller and relying on its onboard rate stabilization PID loop.
DCL is kind of a weird drone racing league since it's made for TV; it's mostly simulator based with, more recently, only few real events a year. The spec DCL drone isn't very capable compared to the more open-specification drones in racing leagues like MultiGP, in large part to keep the events more spectator friendly. This probably makes it more amenable to AI, which is an interesting side effect.
> One of the core new elements of the drone’s AI is the use of a deep neural network that doesn’t send control commands to a traditional human controller, but directly to the motors.
[0] https://www.esa.int/gsp/ACT/
[1] https://www.esa.int/gsp/ACT/projects/rl_vs_imitation_learnin...
> The course design pushed the boundaries of perception-based autonomy—featuring wide gate spacing, irregular lighting, and minimal visual markers. The use of rolling shutter cameras further heightened the difficulty, testing each team’s ability to deliver fast, stable performance under demanding conditions
https://a2rl.io/press-release/9/artificial-intelligence-triu...
Feels kinda similar to the innovation around manned aircraft about 100 years ago when we went from toy/observation platform to killing machine in only a couple of decades. With the ardupilot news today, it was hard to not watch this and imagine the applications to a combat environment.
The optic cable is for the human pilot. An AI piloted drone doesn't need it.
I expect Russia will be ok with it in any situation.
A lot of comments are trying to draw connections to combat drones, but drone racing like this has been a hobby thing for a long time. The capabilities of the drones are set to have an even playing field, not to match combat drones or anything.
These aren't meant to have any parallels to combat drones, drones that fly long distances, or drones that carry payloads.
It's really just a special-purpose hobby thing for flying through a series of gates very quickly. Flight time measured in a couple minutes, no provisions for carrying weight.
I'm afraid not. RC/FPV is already a niche hobby, and media coverage is universally negative. No wonder laypeople mostly think of kamikaze drones when they see something like this.
Yeah, I’m sure this is a great milestone but it isn’t notable until AI is beating MCK[1] who would be the “Lee Sodol of FPV”
Does the jetson board even have the appropriate UARTs on it to talk directly to the ESCs? Your typical hobby grade 5" class size ESC (either 4-in-1 or discrete) cannot talk to two different controllers at the same time. If it's already wired to the four UART outputs from a 30x30 size flight controller (such as something STM32H7 based running betaflight), the ESC cannot be in communication with any other device.
Digging in, you could probably bitbang DShot on Orin GPIOs. It would be really sketchy since it's presumably running Linux on out-of-order cores with giant cache, so there's nothing close to realtime going on, but it's so damn fast that the theoretical problem probably doesn't matter and I bet it would work fine.
Orin also has "safety island" for ASIL applications which has lockstep Cortex-R52s. This would be ideal to run an integrated flight controller RTOS but it's supposedly fused off on consumer boards, lame. There's also a Cortex-R5 called "SPE" that could probably run a flight control RTOS as well.
Regardless of all of this theory craft, I was either totally wrong or accidentally right in two ways, depending on how you want to look at it: according to a sibling post, the STM32 board actually runs the NN-backed stabilization loop rather than a PID one! I've seen this done in research before but never in a meaningfully successful way, so I'm even more impressed now - as far as I can tell, it's _two_ novel solves, an actually working PID-or-better NN-based stabilization system on STM32 _and_ inside-out vision based flight planning for racing on the Nvidia board.
The tech industry is working hard to bring about the Terminator future.
And I remembered The Fourteenth Book of Bokonon, which I had read in its entirety the night before. The Fourteenth Book is entitled, “What Can a Thoughtful Man Hope for Mankind on Earth, Given the Experience of the Past Million Years?”
It doesn’t take long to read The Fourteenth Book. It consists of one word and a period.
This is it:
“Nothing.”
--Kurt Vonnegut, Cat's Cradle
I think we're already deep into large-scale drone warfare. Destroying a third of the enemy heavy bomber fleet is pretty substantial. It feels to me like that attack operated like Pearl Harbor, a marker that the old way of surface naval warfare / air attack was being replaced by a new one.
Don't forget that Russia has their own drones. They were the first to deploy the fiber-optic cable drones as an anti-ECM measure. And of course both sides are ordering parts from China.
Future warfare will be drone Vs drone, and we might see the main battlefield entirely devoid of people until one or other side runs out of drones.
Whole program is about a third of the cost of a Type 26 Frigate.
While I want to reduce the number, I can't but help think how we essentially create them in the first place by destroying the countries which create the economic conditions they flee from.
Extremely affordable!
well, there will be similarly smart "predator"/defense drones. The humans will have no chances on such a battlefield populated by thousands drones per square kilometer fighting each other.
>The tech industry is working hard to bring about the Terminator future.
And i think removing people from the battlefield is a good thing.
>or at least dodge out of where guns are pointing
just a bit of arithmetic comparing new weapons - drones vs. classic guns. Say a radar guided gun takes 1 sec. to train onto a drone and shoot several bullets. The range is max 3 km (an expensive 20mm-30mm autocannon like Pantsir) - 35 seconds for a 200 miles/hour drone. Thus all it takes is maximum 36 such drones coming simultaneously from all the directions to take out that gun. At less than $1000/drone it is many times cheaper than that radar guided gun. (and that without accounting for the drones coming in very low and hiding behind trees, hills, etc and without the first drones interfering with the radar say by dropping a foil chaff clouds, etc.) It is basically a very typical paradigm shift from vertical scaling to horizontal scaling by way of software orchestrated cheap components.
Drones don't remove people from the battlefield, they further the trend of there being no boundary to "the battlefield", putting everyone on it.
They can, depending on how they are employed, reduce the casualties (total and particularly civilian) on both sides of a conflict for any degree of military impact (Ukraine's recent strike against Russian bombers is an example), or they can increase the civilian death toll for marginal military impact (the accounts of Israeli gun- and missile-armed drones directly targeting civilians in Gaza being an example of what that could look like.)
Thus indeed, this made the battlefield larger instead, now common trucks, warehouses and shipping containers are legitimate targets.
What Ukraine destroyed doesn't help either, for example they destroyed early warning airplanes intended to warn Russia if incoming missiles are nuclear or not. How Russia have to assume incoming missiles are nuclear, specially if they are flying in the regions where their land nuke detectors were destroyed too (I think 1 or 2 years ago Ukraine did that).
Thus Ukraine proved, that civilian equipment can destroy nuclear deterrence. Now common trucks and containers are a threat as big as many advanced military hardware out there. A truck with a bunch of drones can open a hole in your nuclear defense as much as stealth planes were needed for this before.
yes, that is the point i've been making for a while - those cheap automated systems, the drones being the first examples of it, is the new MAD/equalizer weapons now available to all countries, not only to the large nuclear ones (and that becomes very important for avoiding future wars giving for example growing doubt that NATO, and USA in particular, would come to the defense of Baltic countries, whereis several millions of drones (including larger long range ones) which Baltic countries can get relatively easy for several billions of dollars may pose an unacceptable high cost to Russia of any potential aggression against those countries).
In this particular case a much smaller Ukraine can use that MAD/equalizer potential to win the war, or at least to get a great negotiating position by systematically severely degrading Russia's strategic capabilities toward making Russia potentially defenseless against US or China, or even say Turkey.
That is how i think they can degrade strategic air/missile defense systems https://news.ycombinator.com/item?id=42529638
And imagine if similarly to the plane attack Ukraine would attack Russian nuclear submarines parked openly at the bases (say using ships for drone launching instead of trucks) - there is no risk of destruction of those submarines, yet 10-50kg drone can damage the skin and outer hull forcing the submarine out of service for prolonged time.
It is very dangerous, since it will mean that an organization with enough drones can dominate society on its own. Much better if humans were battlefield-relevant.
And if you look at Russia your logic does fail on that example - no amount of human losses affect Russia's behavior in the current war as they are sure that Ukraine will run out of soldiers before Russia does. So, from Russia's POV the faster the grinder the sooner their victory.
So you get desertion, refusal to enlist, rapid surrender, and so on. This results in the losing state having to resort to ever more brutal means of conscription such as literally dragging people in off the street, making it illegal to film such actions, making it illegal to leave the country, expanding the age range for conscription, and so on.
That all results in even worse morale which makes your fundamental problems even worse. That, in turn, can motivate the losing nation to expend soldiers/resources on missions which may have some propaganda benefit, but ultimately serve no military purpose whatsoever. And at some point it all just collapses like a house of cards.
---
And I think this fundamental issue of morale will be a perpetual in war. The winner will not be decided by who has the most drones, but by which side's morale breaks first. This is why Afghanistan, in terms of outcomes, is essentially the strongest military nation in the world. They've defeated both the US and the USSR in spite of being orders of magnitude behind in every single measure of military strength - except for morale. Those guys' spirit is simply unbreakable and they will fight you for decades, and to the last man, with absolutely no relenting.
The problem is, in Afghanistan the Western nations didn't do much else than depose the Taliban that took around two years and provide education abilities for women afterwards. But in order to actually achieve change, you have to invest significantly more resources to actually build the foundations for a viable society: democracy, rule of law and an economic perspective for the populace.
In Germany, the Allied Forces stayed for about 45 years, two generations worth of time. Just think of the massive amount of money and resources invested... the first years were taken similarly to Afghanistan - depose the Hxtler regime and rebuild a rule of law afterwards and, as in the Luftbrücke, ensure basic survival. But then, they stayed in for over three decades to make sure that a healthy democracy would not just form but also establish and entrench itself against threats, and that Germany had an industrial base which was used to provide employment and income for the populace. Also, thank God for the Americans deciding not to follow the "Morgenthau plan" that proposed turning Germany into a purely agrarian state with no industrial capability ever again - that would have caused us to follow down the Afghanistan path with utter certainty.
In Afghanistan however, the situation after the immediate war and short post-war period was markedly different. The troops were locked up in their bases outside of bombing jihadists, which meant that local warlords had little to no oversight in their atrocities and stuff like "bacha bazi" (organized child sexual abuse) and slavery went on with effective impunity. The local puppet government barely had any income sources other than foreign aid (and selling opium on the black market) which meant there was no way to form a national identity and storytelling or even a common purpose, and a lack of oversight of the occupying forces over the puppet government led to widespread corruption and looting of the external investments, which led to it losing support across the country. And on top of that, we didn't even do decent oversight over our own troops. Abu Ghuraib is far from the only scandal that was barely prosecuted, not to mention all the other shit that was quietly swept under the rug - that led to the populace despising our troops even more.
We didn't lose Afghanistan because the Taliban are a strong army - they were and are not, just look at the videos from right after the takeover. We lost Afghanistan because we didn't give anyone in the wide population a reason to fight for themselves and not just submit to the next best warlord.
All of the things you're discussing are not things that the US simply didn't bother to try to solve, but we were ultimately powerless to do so. Americans would never tolerate US soldiers dying by the tens to hundreds of thousands as would have happened if we actually tried to enforce order on foot. So we were left with proxy soldiers, contractors, and a money printing machine. But that simply wasn't enough to defeat the Taliban, let alone carry out the grand changes you mention.
So yeah as morale collapses you're left expanding the age of conscription, which further collapses morale. There were plenty of sardonic jokes about the Volkssturm, 'the people's brigade.' Why is the Volkssturm the state's most valuable resource? Because they have silver in their hair, gold in their teeth, and lead in their bones.
[1] - https://en.wikipedia.org/wiki/Befehlsnotstand#Nazi_Germany
Battlefields have the inconvenient property of sometimes coming to where you are. Even if you would rather not participate in any way.
Currently in democratic countries one of the brakes on war is that you need "boots" on the ground, and "boots" on the ground results in caskets draped in flags on TV. Which result in people not voting for you come next election. If you don't need humans to fight on the ground anymore (or you can get away with drastically fewer humans on the battlefield) then you will get a lot more war, and a lot more battlefields in a lot more places.
That's the problem. First order effect is of course good for the humans who don't need to die on the battlefield to achieve some goals. Second order effect is what I'm worried about. The lot more suffering caused by a lot more wars and battlefields in a less stable world.
And that is assuming you need the resources of a state to fight these autonomous wars. If the tech is cheap enough, and hard to "control" enough that it is available for organised crime you might see it used in assassinations, gang warfare, and protection rackets. And then we all will live on battlefields. Third order effects are the people hurt by the anti-drone weapons missing their target or activating the wrong time. Fourth order effects are all the constraints and weird technology restrictions they will put on tech trying to stop the proliferation of autonomous drones.
I'm still concerned about the worst case scenario there being Microsoft getting their trusted computing wet dream. However I'm hopeful that it ends up being nothing more than embedded microscopic serial numbers in anything resembling a microcontroller, ID requirements to place purchases, and legal requirements for disposal (no more garage sale electronics).
The other scenario that I think might be more likely is radar and video surveillance covering every inch of every city and a related domestic agency capable of fielding rapid, tightly targeted anti-aircraft measures on a large scale.
Here in Sweden we instituted mandatory military service we did so because we wanted to ensure that there was no military class that if they decide to can take over. We knew the cost, and the cost is worth it.
In normal times the cost is simply to do ones mandatory military service.
This protects against coups, ensures your power in society and prevents groups of officers and soldiers etc. from taking over.
Meanwhile Spain suffered an attempted coup in 1981 [0] while mandatory military service was still in place [1]. The conscripts did not play a role in protecting democracy.
[0] https://en.m.wikipedia.org/wiki/1981_Spanish_coup_attempt
[1] https://en.m.wikipedia.org/wiki/Military_Service_(Spain)
In both cases I very much doubt that people lacking the training, organisation and weaponry of the professional military will be able to beat them in contemporary circumstances.
I realise military service means people have some training, but as much as the professionals? What about air cover, heavy weaponry, communications? What about timing - a coup might be over before conscripts can react.
Most of all, is there historical evidence this works?
In actual coups, it's often a small cadre of well-connected higher officers who do the work. It's not the whole military. By the time the whole military (or country) realises what's happened, it's already happened and there's not a lot they can do.
This is the first I hear that this would be the motivation.
The main motivation is that for a small country like Sweden to have enough manpower to defend itself adequately, conscription is necessary.
I agree with your other points, but this only helps with (physically) extending the battlefield, at least going by the current war in Ukraine. It's not only the line of contact that is now part of the battlefield, there's also a band of 10-15 kilometres (if not more) on each side which is now part of the active battlefield because of the use of drones.
Even though I have to admit that it looks like the very big power asymmetry in favour of cheap drones over almost everything that moves down bellow (from mere soldiers on foot to armoured vehicles) has helped with actually decreasing the number of total casualties (just one of the many paradoxes of war), as it is now way too risky to get out in the open so soldiers do it way less compared with the pre-drone era.
You're mistaking the removal of certain soldiers for "removing people". There will absolutely be people in future battle fields, mainly civilians, or as we call them now, terrorists.
Off the shelf parts are available for 200mph quads (that’s without payload though)
See paragraph 87 by searching for "THE MOTIVES OF SCIENTISTS"
https://dn790000.ca.archive.org/0/items/IndustrialSocietyAnd...
"92. Thus science marches on blindly, without regard to the real welfare of the human race or to any other standard, obedient only to the psychological needs of the scientists and of the government officials and corporation executives who provide the funds for research."
Politicians will never go outside again. The only defense is to be loved by all and have no enemies. Or, the more likely scenario: disguises and full anonymity.
We'll simply never know who we're being ruled by.
Of course deeper into Russia that's safe .. but instead you have the problem of a huge area to cover. You can protect a few high value targets but not everywhere. Consider something like the early stages of the Iraq war: target every single civilian electrical substation and petrol station with a drone bombing.
ramp up power levels so dwell time might only be 1/2 second? maybe. but then there is a race for rapid target discrimination. and then ablative armor on the drones (cheap and easy to 3d print), and backup cameras, etc.
https://www.darpa.mil/research/programs/offensive-swarm-enab...
I do think we will see more and more laser systems installed on large naval vessels as an alternative to close-in weapon system (CIWS), which, in practice, has not been very effective against missile attacks.
The way air-air missiles typically work is massive fragmentation warheads and proximity detonation. Even then, it doesn’t always work.
The pro however is that the interceptor can position itself somewhat arbitrarily, and there doesn’t need to be ongoing line of sight from the initial detection point. In some cases (theatre radar systems, standoff radar systems), they may not even need to be able to see the target drone until they’re within detonation range. (Think ‘hides behind a random bush until they’re within detonation range of the target drone as it tries to zip by, then explodes with no warning’).
Then of course the attacking drones will add randomness to their attack patterns.
Lasers have the advantage that it’s essentially impossible to dodge a laser beam (speed of light meaning that for objects within the ranges we’re talking about are lazed the moment the laser turns on), so it’s purely an aiming/detection/line of sight issue. But they have the disadvantage that if the drone gets out of the line of sight, it’s completely ineffective. So they’ll need to be placed at locations that overlook large areas and have good visibility towards the entire potential approach area, which makes them vulnerable to artillery, massed attacks, etc.
Well, at least we know what shape the new arms race is taking!
The advantage of course is that the interceptor can position itself arbitrarily, and there doesn’t need to be ongoing line of sight from the initial detection point.
Lasers have the advantage that it’s essentially impossible to dodge a laser beam (speed of light meaning that for objects within the ranges we’re talking about are lazed the moment the laser turns on), so it’s purely an aiming/detection/line of sight issue. But they have the disadvantage that if the drone gets out of the line of sight, it’s completely ineffective.
Well, at least we know what shape the new arms race is taking!
Various entities related to the US military have been simulating and physically testing various approaches to that for years. There are blurbs in the press here and there. In some cases source code even got published (not clear if that was intentional though).
Two opposing Iron Beam lasers pointed at the allied one?
But also, low-flying drones using ground cover to get close, and in the final approach attack from multiple angles at the same time so it physically cannot rotate fast enough to reach all targets before impact — if it has a maximum 180° rotation time of 0.2 seconds, and the attacking drone has a terminal speed of 100m/s (~224 mph), the last 20 meters has to be clear of all cover, including the soldiers who are operating the Iron Beam and might find their bodies being kept between the drones and the laser itself until the last moment.
Of course more drones works. But more drones here is less drones there. It means lasers are an effective deterrent against opportunistic attacks.
I am fucking terrified of drones.
Increasingly we are also seeing a world where the technology to shape a cultures mind share can be deployed with a few dozen lines of code and a malware bot net rather than a sophisticated and well funded mass media operation a la the 1960s western cultural revolution supported in part by the CIA. You don't even need to blow up the enemies country, you can convince them it is in their best interest to be subjugated and they will remove their own naysayer internally and roll out the carpet for you when you arrive and proclaim your regional Obergruppenführer to meet production quotas.
We might be able to put a pin in this tech from a policy perspective, but the cat is way out of the bag as far as the tech goes. A cell phone already has all of the sensors you need baked right into it (honestly, we can thank mobile devices for getting the cost down). An ESC for a motor is a cheap microcontroller and a couple of MOSFETs. The frames can be made of cheap plastic. Even if things like ArduPilot didn't exist, a smart EE student could build one from scratch, including the flight control software, using parts from Digikey and relatively basic PID control code.
The cat is definitely out of the bag.
Nice.
America insists on making sure that guns are universally available so that school shootings can still happen. Doesn't register. The death toll seems to be politically acceptable.
Plus, you got a cool and potentially lucrative hobby, designing exterminator machines. Why bother with children at that point?
There are much, much better targets to be had.
Your point on the dwindling barrier to implementation stands.
the premise is that the person doing it is very mentally ill. the question, "why would they do that when they could do something else that makes more sense?", doesn't make a lot of sense itself under the premise.
The hard part is that there is no effective way to regulate anything in the supply chain involved except for the explosives themselves. Everything else is super commoditized at this point and, other than the props, very multi-purpose. The first significant hexcopter I built used a BeagleBone Blue for processing, generic ESCs and BLDCs for the motors, and an aluminum frame that I cut out of aluminum tubes from Home Depot. Max takeoff weight was 55lb, because that’s the heaviest it could legally take off with. This was 7 years ago.
one thing in societies favor though - sufficiently unstable lunatics tend to self delete themselves in various ways by being unstable lunatics. few tend to be in the “sweet” spot of dangerous lunatics who are stable and focused enough to follow through successfully with a dangerous plan. thankfully.
For example - most people who could synthesize multi-kilo quantities of TATP without blowing themselves up and successfully build a DIY drone to carry it have better and more productive things to do with their lives. at least in the west.
There is no Jedi Council to appeal to, no wise group of non-aggressive nations gathering to pacify the troublemakers.
A large number of front-line FPV drones are equipped with automated last-second targeting systems like https://www.youtube.com/watch?v=coUwYOyIoAU , based on Chinese NPU IP / CCTV systems and readily available as full solutions on Aliexpress. The basic idea is that if the drone loses control or video link due to EW countermeasures, it can continue to the last target.
Loitering and long-range fixed wing reconnaissance drones have been fully autonomous since the beginning. One common recent technique taken from traditional "big" militaries is the use of loitering autonomous high altitude base stations with Starlink or LTE on them providing coverage to the battlefield below, since it's much harder to jam things when they are flying high above the ground.
Rather, it is you who does not know what you are talking about. Here is a real frontline video characterizing these systems. Yes, it is all still analog FPV. The lock-on system selects a target and overlays the reticle on the analog video. As the FPV flies closer and encounters the jamming from the target, the lock-on unit ensures it is still a hit.
These have fallen out of favor as fiber optic is a little easier to get than it used to be but they are still in wide use.
> Технологія нова, потребує вдосконалення і масштабування.
> Eng: "The technology is new and will need improvement and scaling"
I don't understand what you're trying to prove. They do exist and I never said they don't. They keep popping up here and there, mainly working in demo conditions against static contrast targets.
My point was that so far, these things are just curiosities with very limited usage and there's no mass adoption. Maybe some, but there's some of everything in this war. All the main uav units that I'm aware of use manually controlled fpvs and there are reasons for that.
> These have fallen out of favor as fiber optic is a little easier
Oh gosh.
Make that point, then! Nothing in your original comment suggested this, just hostile dismissal.
Now that you’ve written a more substantive comment I think we actually agree overall. Most operations in the Ukraine-Russia war are manual piloting. Autonomy is over-hyped overall so far. However! A large number of autonomous systems have still been deployed and interest in autonomy is only growing. Both things can be true at the same time.
> Oh gosh.
Come on, read the whole sentence please. Lock on targeting modules are absolutely being superseded by fiber optic as it becomes “easier” to acquire than it used to be.
https://understandingwar.org/backgrounder/battlefield-ai-rev... was posted by a sibling commenter and is a fairly accurate summary to my knowledge, including a substantiation of the notion that depending on how you look at it, lock on modules were a stop-gap before fiber became available or fiber is a stop-gap before good autonomy becomes widespread.
Governments are falling over themselves to: acquire drones, figure out how to defend against existing and future drones, and to figure out how to exploit them well. Given the recent attack against Russian bombers, I find it hard to take you seriously here.
Hell, the US knows it can't compete with China on aircraft numbers, and is placing its money on collaborative combat aircraft to give it the advantage. That's about as strong an endorsement as you can get.
But then they go and say "drone swarms will defeat all future adversaries!"
Like in the Ukrainian context everyone seems to think the drone swarm was the deciding factor and is saying "this will replace air forces!"...kind of ignoring the multi month infiltration and espionage operation which got those systems in range (they were literally trucked right up to almost the fence line).
Many/most folks use the term "drone" to talk about CCA's and other expensive platforms. In fact, "drone warfare" predates the common application to quadcopters, people were calling the Predator drone a drone in the early 2000's. I do agree that calling everything a drone is annoying though, and makes it hard to know what people are talking about. "AI" is having the same problem today.
But every post vaguely about drones on HN has a whole bunch of people acting like a 1000 quadcopters will replace an F-35.
I don’t really put much stock in your average person’s take on warfare generally, so I’m not too bothered by the torrent of misunderstanding. You see the same thing with AI/AGI, and much of it is fueled by those garnering for clicks.
I will say I missed in my original response that the OP was taking exclusively about autonomous devices, and in that case I would agree with their take.
These bombers attacks were done with manual control too. These drones had LTE modems and on footage it's clearly visible that they controlled by operator.
People can't read these days, especially if it doesn't match the reality they build in their heads.
I'll skip the shitty retort about not reading.
One of the theories for why there were tires on top of the russian planes that were bombed is that it confuses automatic targeting systems by breaking up the profile of the airplane used in automatic target recognition systems.
Hell, even hobbyist level DIY drone stuff can be easily programmed to run an autonomous route with or without a radio link connection. This is a huge reason that GPS is just constantly jammed in this part of the world. If you can get a GPS signal on the battlefield, you can tell a drone to go destroy something.
> Hell, even hobbyist level DIY drone stuff can be easily programmed
Lock on a moving target and hit it is not the same as put waypoints in INAV. My point was that there's still no mass adoption of target locking or self-aiming drones, overwhelming majority of hits, on both sides, are done with regular FPV drones with very standard school hardware that's barely modified for combat use (namely: custom frequencies for VTX and ERLS).
As long as you define ‘drone’ as a tiny quadrotor. Missiles like Sidewinder and Hellfire, cruise missiles like Tomahwak, fire-and-forget MANPADs, GPS-guided gravity bombs, even ICBMs with MIRV warheads. All autonomously travel to their target and destroy it.
There are even some loitering anti-tank missiles that climb up above the launching aircraft and sit on a parachute for a while until they see a tank to destroy. The pilot never has to see the tank.
All autonomous and adopted.
The main novelty in the electric drone tech is very very low cost.
Why is this, for the rest of us?
There was order from the higher ups: protect the planes. There was no specific order like "build a garages for planes". So they put tires on planes and called it protection. Now soldiers have to move tires around, fill the journals, sign them up. New procedures are developed. Order is fulfilled. Everyone was happy. May be someone even got a promotion for creating a plane defence system so cheap.
Of course, those have significantly less performance than the one you put in an airliner or ballistic missile.
As you mention yourself, its a question of good enough. You need to be a lot more accuracy to hit a city after a twenty minute sub-orbital coast, than to find the nearby trench. And yes, computer vision is used to correct for drift.
[1]: IMU's on DigiKey: https://www.digikey.com/en/products/filter/motion-sensors/im...
"Promises of an immediate AI/ML drone revolution are premature as of June 2025, given that both Russian and Ukrainian forces will need to allocate more time, testing, and investment to deploy these drones on the frontlines en masse. Russia and Ukraine will continue improving their ML and machine vision capabilities while training and testing AI capabilities. Russia and Ukraine will then need to tackle the issue of scaling the production of the new AI/ML drones that will require additional time and resources to facilitate. Russia and Ukraine may start to use some AI/ML drones to carry out specific tasks in the meantime, such as striking certain types of targets like armored equipment or aircraft, before learning to fully operate on the battlefield. AI/ML drones are also unlikely to fully replace the need for the mass of tactical FPV drones over the coming months because the latter are cheaper to produce and adapt to the current battlefield conditions at the current state of technology."
[1] https://understandingwar.org/backgrounder/battlefield-ai-rev...
Anti-anti-drone avoidance systems on Russian zala's is the only example of autonomous action that I can remember.
Update: I'm not saying people shouldn't develop this, we're never going to squash human curiosity. But when I see this kind of stuff, I'm deeply troubled by how bad actors (state and non-state) will use this.
I hope our security services are working hard on countering these potential threats.
I'm having a hard time believing this is effective.
> The only reason we don't use them now is due to all the collateral damage
Russians don't care about collateral damage and there doesn't seem to be any evidence of them using such weapons?
It would be devastating in the local battlefield, potentially frying radio or other equipment depending on the size of the device or how close you could lob it towards the enemy before going off; but with the low wattages many non-military communication devices use today you would also be blasting horrible noise to all of them beyond the local area and disrupting communications across potentially multiple neutral countries.
It would be a large act of aggression against any countries around them and NATO, and at scale possibly even piss off far away countries like the US and China. Especially large EMP devices could even be temporarily misidentified as a nuclear explosion and gain the immediate full attention by any nuclear powers watching out for it.
For undirected EMPs, consider that a lightning strike is in the region of about 5 gigajoules, and the difficulty of pumping that much energy through a coil.
My expectation is that a CHAMP warhead (not missile, warhead) would have a EMP range of something under 300m (probably less), even then vs. unshielded electronics (compare the effect on your body of being in a car that's struck by lightning vs. being on a 100m away on an open field that's struck by lightning), and only even that much by having a high gain (highly directional) antenna: https://en.wikipedia.org/wiki/Counter-electronics_High_Power...
That said:
> but with the low wattages many non-military communication devices use today you would also be blasting horrible noise to all of them beyond the local area and disrupting communications across potentially multiple neutral countries.
This doesn't need an EMP, it's "just" jamming, and Russia (amongst others) already does this: https://gpsjam.org/?lat=50.32113&lon=41.41602&z=3.0&date=202...
Yes, a wall of $150,000 bushmasters with some servos can take hurl enough lead in the air to protect a city from a single gulf-war era $100,000 lawnmower engine on two meter wings bumbling in a straight line at jogging pace.
We’re currently in the “a shipping container on a semi can could launch dozens of $2000 racing quads with molotov cocktails zip-tied to the bottom with enough agility to thread a needle faster than a turret can swing its own mass”
And the writing is on the wall for some near-future “any nation state could drop sci-fi cluster bombs that shed ten-thousand 250gram racing quads that can overwhelm even the most advanced point defence just by numbers and it’ll be cheaper than a conventional 2000lb bomb”
If it has a range of 1.5km, then the drone needs to be moving really fast in order to move a couple of degrees per second.
If it's going like 80mph, then that's 36m/s, which comes out to <1.4 degrees/s at 1500m away. For 1000m, you get ~2 degrees/s. Not to mention that this velocity is at a right angle to the turret, and you have to close the distance.
Firefighters use them to search for missing persons but also to get aerial images and a better overview of larger scenes as "running around" is often not possible or doesn't help that much with the overview.
Police is using them to take pictures of accidents. It's easier to see tire marks and the whole "history" of an accident from above. Really reduces their time on a scenery to take pictures of everything.
Cons: massive invasion of privacy and probably illegal.
Pros: looks cool.
https://www.youtube.com/watch?v=yz2in2eFATE
The speed and flawlessness is quite impressive considering it is being resolved with what I imagine is noisy inertial data and a motion blurred CCD camera.
It's really cool to see this happening fully autonomously and at such high speed. I wonder if the use of AI means that the approach is fundamentally different, or if it uses the same principle of minimizing snap?
https://www.ted.com/talks/vijay_kumar_robots_that_fly_and_co...
> It allows weak players to have more leverage
I think it depends on the dominant type of defensive counter-technology used. If it's something with high capital costs like a laser or a microwave, then it will centralize combat because the USA could invest in the infrastructure needed to defeat terrorists, but not the reverse. On the other hand, if you can effectively destroy these things with birdshot, then it may not be a problem for humanity at large. I could imagine you could make some device that tracks the drones and shoots them down for much less than the price of a drone.
For example look at the iron dome: it is effective defense measure but very difficult to scale up to counter ICBMs.
The iranian icbms from Yemen gets shot down by other systems. I don't think they are scaled up versions of the iron dome. What's interesting is the lack of confidence israel has in predicting their trajectory. They usually send up multiple interceptors for a single missile and put sirens on for half the country.
The only really economical counter i've heard of is lasers but it doesn't look like they are coming any time soon.
I've got no idea what method Israel uses to counter drones, but they certainly have struggled with them.
My absolute nightmare scenario is Iran via its proxies unleashing swarms of autonomous kamikaze drones in population centres.
The cost of a FPV quadcopter is pennies on the dollar cost of a defense system. You can 3D print a dozen frames in hours with a decent printer. Its just bolt on the parts and do a quick function test with a script or controller and its ready for action.
It unfortunately goes even deeper than that. The Quadcopter FPV community is watching their open source software actively be picked up for warfare, and can often tell what version they were running when watching released Russian & Ukrainian footage later.
Every beneficial step we make in the maker community will be used to expedite death in conflict. A few 3D printers and a digikey order are all you really need to seed an insurgent movement at this point.
I’m really curious how this would perform in messier, less controlled environments.
Usain Bolt was the fastest human sprinter in the world, but compared to a good motorcycle over paved road he's obviously not very fast, and likewise compared to an emu. Nevertheless, Bolt's 100m performance drew big crowds, even though people also watch Motorcycle racing and (I think?) Emu racing.
It's like speed running, the categories are arbitrary and self-selecting. Why the Modern Pentathlon? Why not. Why Super Mario Warpless? Why not. If everybody wanted to do Super Mario, only the odd numbered levels and also you must kill all the enemies, that's what the run is, our choices are arbitrary and we value whatever we like.
https://www.npr.org/2023/08/30/1196777528/an-ai-quadcopter-h...
or was it overfitted to this specific course?
Anyway given the state of the art, flying autonomously at great speeds and beating human champions without pre-training, i.e. on an unknown race track, would be a much bigger breakthrough than just beating some human champions (which has already happened except in a less official environment). You can rest assured that if that was what the team achieved, the article would be telling us all about it.
That being said, I'm sure they have a base model too, so I'm right back to wondering about the parent question: would it work if you set it down in front of a few fresh gates?
Why Generalization in RL is Difficult: Epistemic POMDPs and Implicit Partial Observability
https://arxiv.org/abs/2107.06277
OK, it's a robotic zookeeper looking for the otter cage.
Where does it say they had no prior knowledge before the event? I can't find that in the text. Is it in the video?
I guess there's no paper yet.
Reading back through it, I'm synthesizing this statement. It's never said explicitly, and I could very well be wrong.
I'm combining the knowledge that the novel development here is that the event supervisors control the track with the fact they're showing off a training run in their video.[0] The video also links a few papers from the teams past that have some additional clues.[1][2]
> The reason for this mostly lies in the real-world aspects of the competitions. They take place in environments previously unknown by the teams, with no opportunity for benign, solution-specific changes, and little time for adapting the developed solution to the environment in situ. Moreover, competitions often pose a more challenging environment, with gates located slightly differently than on the precommunicated maps or even moving during the race, unforeseen lighting effects optimized for spectators rather than for drones, and large crowds of moving people around the flight arena.
This makes it sound like they're at least given the layout.
Note this was from a different competition (Artificial Intelligence Robotic Racing by Lockheed, with DRL) back in 2019. The other paper is from 2024, but I don't have access.
https://www.youtube.com/watch?v=yz2in2eFATE
Yeah, I read it as saying they were given a short amount of time to train before the race, where they wouldn't have time restrictions otherwise. Which makes the result more impressive, of course.
Let's say your task is to move a human from A to B (by a pre-designated route) as fast as possible. The only conditions are vehicle weight, no outside radio contact and no damaging the road (assume each vehicle goes separately, so e.g. slip stream effects don't matter). You can rely on the human to drive or use AI, you can go on the ground or fly through the air, anything is allowed. What would be the best way to do this?
Desiccate the human and compact him into an aerodynamic shape. Carry the (now much lighter and more aerodynamic) human inside a small rocket.
https://www.twz.com/news-features/ukrainian-commanders-exclu...
1. I didn't see it stated explicitly, but I presume the neural net is on the far end of a radio link somewhere, not running on hardware physically mounted on the drone?
2. After viewing the FPV video on the linked page: how the hell do human pilots even come close to this pace? Insane (even assuming that the video they're seeing is higher quality than what's shown on YouTube – is it?)
3. The control software has access to an IMU. This seems to represent some degree of unfair advantage? I presume the human pilots don't have that – unless the IMU data is somehow overlaid onto their FPV view (but even then, I can't imagine how much practice would be needed to learn to make use of that in realtime).
2) No, the video the pilot sees is usually quite bad. Racing pilots usually use either HDZero (mid resolution video with weird pixel artifacts sometimes) or analog video (looks like a broken 1980s VCR). It’s amazing what they can fly through. These DCP spec drones are also slow by racing standards. Look up MultiGP racing, it’s even faster.
3) It can be overlaid but it’s useless. The human pilot is using the control sticks as the input to an outer rate regulation loop which contains the gyro as input to an inner stabilization loop though, so the IMU is still in the mix for human control.
2. The video they're seeing is worse. Spectators typically see the frames saved directly from the camera, but the pilot will be seeing them compressed and beamed over the air to their headset. See vid.
3. The human pilots do actually have access to it. Not directly, but the flight controller translates their inputs and makes use of the IMU to do so.
I’m reminded of when the US military figured out it should just replace all its proprietary field drone controllers with Xbox controllers because every single grunt that enlisted already had 10,000 hours on the things. If the future of warfare is drones, Christ, that video is terrifying.
1. Less precise. Gimble size matters.
2. All inputs sprung. This is exactly what you want for your three rotational axis, but you absolutely do not want your throttle resetting to 50% when you lay off. You can fix this using 3D mode where the zero setting is in the middle, but then you lose even more precision.
3. Circular inputs. This means at low or high throttle you have less roll available.
The main reason you'd want a gamepad is the size and shape. They do make gamepad-style radios, like the Radiomaster Pocket, which combine the best of both worlds.
You can pick up a simulator for $10-20 if anyone wants to give it a whirl, and many are even on Steam, but the general recommendation is to pick up a dedicated radio as soon as possible.
Note that this mainly applies to FPV quadcopters, due to how sensitive and twitchy they can be. When it comes to controlling pretty much anything else (I'd argue even most planes) these advantages are no longer relevant.
Since I can't access.
Ah yes. No mention of the real big use case
...killing you, their target.
I'm mildly suprised that the US hasn't seen a breakout of car bombs since Oklahoma City or WTC. It seems that the tradition force of using guns for the frequent mass casualty suicide terrorism events is too strong.
I got out of doing drone work because of all the FAA restrictions on where you can fly drones now. Within 30 miles of a major metro area? Nope. Within 20 miles of an airport? Nope. I'm exaggerating of course, but it got to a point where I was having real problems trying to find areas where you can fly a drone just for fun so I just gave up and quit.
My more immediate fear would be how the gov can control who and where these drones will be able to fly. If some revolutionary built a swarm of drones, it would be pretty easy (I would think) for the gov to just jam the signal and shut them down.
The parts? I'm not worried about. Its the gov holding the keys to access that makes me more worried.
The Black Eye is a suitcase-sized radio noisemaker that can muddle the signals that control all but the best fiber-optic drones. According to an electronic warfare expert who writes under the pseudonym “Roy,” Black Eye can ground surveillance and attack drones from as far as 4 kilometers away, “when located high enough.”
Unlike many other jammers, which target the drone, the Black Eye targets the drone’s operator—blocking a drone’s command signal at its source. The new jammer “is appearing across the whole front,” Roy wrote. “This is a serious development for Ukraine.”
https://euromaidanpress.com/2025/05/06/blackeyejammers/
More recently:
Kvertus’ innovative EW backpack system provides Ukrainian troops with a mobile counter-drone shield for just $7,000. Operating in the 720-1050 MHz range, it jams and disables threatening Russian drones.
https://euromaidanpress.com/2024/05/17/the-ew-backpack-revol...
There is no signal, and there is no operator.
The Department is mitigating the potential negative effects of unmanned systems on U.S. forces, assets, and installations – at home and abroad. A critical portion of our efforts, particularly in the near-term, comes from improving our defenses, with an emphasis on detection as well as active and passive defenses
https://media.defense.gov/2024/Dec/05/2003599149/-1/-1/0/FAC...
No gps, no fiber, no 5g, no jamming except microwaves. A python file and a target.
Scary times ahead.
and better guidance software. Yeah, there's a lot of room for improvement
"Traditional waypoint navigation assumes movement through a series of Cartesian positions. But in pursuit dynamics, for example, what matters is directional alignment over time"
https://github.com/VoxleOne/SpinStep/blob/main/docs/01-ratio...