Top
Best
New

Posted by _tk_ 6 days ago

I fought in Ukraine and here's why FPV drones kind of suck(warontherocks.com)
197 points | 338 commentspage 3
103e 6 days ago|
Don't drones have another advantage not mentioned here -- counter-battery against operators being more challenging?
originalvichy 6 days ago||
Drone pilots regularly die due to the source of wireless signals being found. Especially in the built up areas where they cannot operate from a trench or bunker. It is a challenge but there have been methods for this for a while now and it has shown. Even recently there’s been reporting that priorities of some drone teams are now purely anti-pilot activities compared to other targets.
103e 6 days ago|||
True - and I hope I didn't give the impression that drone operators aren't taking significant personal risk, but compared to the alternatives for short range indirect fire (mortars) it seems like these systems should be less vulnerable?
throwawayffffas 6 days ago|||
Fiber wires are now the standard for most low flying drones.
throwawayffffas 6 days ago||
More challenging than what?
orthoxerox 6 days ago||
More challenging than counter-battery fire against artillery, which is basically a solved problem in warfare.
throwawayffffas 6 days ago||
But not more challenging than counter battery against teams firing Javelins or other portable anti tank weapons.Or teams using Switchblades.

FPVs are man portable guided munitions, not artillery. Pretty much all existing man portable guided anti tank weapons are better than FPVs at their job.

And artillery is better than any of them at it's job. While FPVs can score kills they have minimal suppression effects, when an FPV hits a friendly, everyone else is going to keep moving, because stopping will offer them no benefit from the next one, and the next one might be minutes out. When an artillery round lands everyone hits the deck.

103e 6 days ago|||
FPVs don't seem anti-tank replacement -- they do seem to have a role against soft targets ie against massing infantry, c2 nodes or suppression of enemy mortars. In this role, from a distance, they seem harder to suppress than the alternative, ie mortars.

Also these are immature tech... I suspect at least some of the issues identified will be mitigated in time.

koonsolo 6 days ago|||
> Pretty much all existing man portable guided anti tank weapons are better than FPVs at their job.

Sure, but a Javelin missile costs more than $200K. You can have 200 fpv drones for that price.

throwawayffffas 6 days ago||
Yeah, but it takes like two guys to carry and use the javelin. 200 fpv drones need like a company to be deployed.
aubanel 6 days ago||
One of the key points of the article is "I feel FPV drones to be mostly a failure because their success rate is low" Why is that a failure? If one 500$ drone has even only 10% success rate, if the target is a 1M$ equipment it's still a win!
randomNumber7 6 days ago||
It is only a matter of time until those drones fly into their target fully autonomously with machine learning.

Heck, I could build that with hugginface (I will never do that) in a few evenings if you are ok to blow up the wrong target with a single digit percentage.

euroderf 3 days ago|
So the degree of collateral damage will depend upon your AI provider.
xg15 6 days ago||
> Drones also operate in a cluttered segment of the electromagnetic spectrum. First-person view drones use unencrypted analog radio signals, and in hot parts of the front, as many as a dozen drone teams may be competing for use of a handful of frequencies (a consequence of using cheaper components). This results in the need for sophisticated de-confliction procedures that, quite simply, do not always work. Even when de-confliction works, sometimes a team must wait as long as half an hour for a frequency to become available before takeoff. If it does not work and two drones find themselves in the air on the same channel at the same time, they will interfere with each other’s signals, usually resulting in a crash. On top of that, the enemy’s drones also fly on the same frequencies, which can also result in interference and a crash.

This gave me the somewhat macabre image of Ukrainian and Russian drones doing automated frequency coordination with each other, so they can orderly proceed in bombing each other's soldiers.

I don't think that's what happens though. But I'm surprised flying drones in the same area as enemy operators is even possible. Wouldn't both sides try to jam or take over each other's signals, deliberately blocking channels, etc - so that in the end, no one could control anything?

Or, if the signals are really unencrypted, what keeps anyone from setting up a radio beacon that just spams the "detonate now" signal on all channels at maximum power. Instant drone-free zone?

dzhiurgis 6 days ago|
I suspect jamming and spoofing isn't that trivial and requires lots of gear that needs to be on while giving out its location...
Cockbrand 6 days ago||
I guess I'm missing something, but why isn't the problem of finicky steering solved by adding auto-stabilizing software? Would that take away too much of the maneuverability?
koonsolo 6 days ago||
My first reaction was the same. I have a small indoor quadcopter with 2 main modes: freestyle (like a helicopter), and an easier mode that keeps the drone hovering.

My first thought was, why not use the easier mode (press forward to go forward, back to go back, etc.)? But looking at those war videos, these drones always come at an angle towards the target. And in that sense, it's easier to use the more difficult helicopter mode. What I mean is, once you know the helicopter mode, it's easier to do this kind of maneuver than using the "easy mode".

tguvot 6 days ago||
i saw ukrainian footage of drones where they switch to/from auto hover mode
FridayoLeary 6 days ago||
The author offers no major new insights on the effectiveness of drones. His counter argument is against the maximalists who claim quadcopters are revolutionising warfare and render armour and artillery obsolete. But nobody serious ever suggested that. The same way tanks are relevant despite rpgs. They simply represent a new element in the battlefield, and a useful one as well. The fact that they are not ideally suited for dropping bombs doesn't matter. They are great for surveillance and giving units situational awareness, and the fact that they can occasionally be used to attack targets that otherwise would be impossible simply augments their usefulness. The article is interesting, but it's attacking a straw man. I have a great respect for the ukrainian armed forces but to be perfectly honest their combat effectiveness is not exactly world beating. The suggestion that NATO should be taking lessons from how Ukraine is fighting Russia is odd.
GiorgioG 6 days ago||
FPV drones don’t suck if you know what you’re doing. If you don’t have proper training, you’re going to suck at it.
dewey 6 days ago|
Sometimes you don’t have time and resources to train everyone to be great. Good enough will have to do.

> As a result, training a highly proficient operator can take months. A standard, base-level course for Ukrainian drone pilots takes about five weeks

GiorgioG 6 days ago||
No disagreement there, but the title states "FPV Drones Kind of Suck".
msgodel 6 days ago||
Soon they'll be using CV and won't need FPV.
palata 6 days ago||
You still want someone to get the drone to a place where it can see the target, and someone to select this target.

Only then can CV do the last part ("terminal engagement"). But that also means it won't go inside a hangar and find the target there.

bluGill 6 days ago||
IF you tell it go into the hanger and find a target CV can do that. It might not be the best target in the hanger, but that doesn't matter too much if you can get in.
palata 6 days ago||
If you add enough "if's", then surely everything is possible :-).

I don't think we're anywhere near having drones that happily fly above a war zone, detect an interesting hangar, find a way to get inside and select a target inside.

Currently they mostly fly FPV drones manually. The next basic step is to have "terminal engagement", where at some point they can select a target and the drone will fly autonomously to it using CV. But in order to do that, the drone will need processing power, and therefore it won't cost 500$ anymore.

Would you rather go for a drone that costs 5k and can use CV for terminal engagement, or 10 drones that cost 500 and simply stay on their latest vector if no command arrives?

05 6 days ago|||
Terminal guidance boards are already available on Aliexpress for $300-ish. Using YOLOv7 no less. The future is now.
palata 6 days ago|||
Are you telling me that you can search for "terminal guidance" on AliExpress and find such boards?
msgodel 6 days ago|||
Yeah I think people overestimate how big yolo models are and how much compute costs.
palata 6 days ago||
The drone is 500$. If the board is 300$, it makes it a lot more expensive. What's the value of that board? How much does it help? Remember that you need to fly it with FPV to the place where the operator can select the target. So you have a radio link at some point, that can be jammed.
msgodel 5 days ago||
He's being generous. This could be done with a raspberry pi.
palata 5 days ago||
Have you tried it, or are you just guessing?
msgodel 5 days ago||
Yeah I've run various yolo implementations on my main computer and I've read what other people have measured on current and last gen raspberry pis.

Again I think people overestimate how much compute this takes and underestimate how cheap embedded compute has gotten.

palata 4 days ago||
I see on the rpi website [1] that an RPi 5 costs 120$. That's 25% of the price of the FPV drone (500$).

Then you have to integrate that RPi to the drone, both mechanically and software-wise. From YOLO, you need to actually get into commands for the autopilot. There is work involved.

And another question is: can you order millions of RPis like this? If not, what is the cost of alternatives?

Not saying it's impossible, just saying that people tend to underestimate those questions. I know it from experience, I have seen projects fail because of that.

[1]: https://www.raspberrypi.com/products/raspberry-pi-5/

bluGill 6 days ago||||
I was trying to state you fly it manually to the hanger. Once it locks onto the hanger (that is not your own hanger) it can fly - even if cv messes up we are at a target so anything destroyd is okay even if not what you want.
palata 6 days ago||
Oh, I had misunderstood it. But still, it's a lot harder to do that succesfully than to lock on a target and fly straight into it. Which already requires the compute power to do it, which makes it a lot more expensive than it is without.
bluGill 5 days ago||
Sure, but we can trade off computer power. Pi level computers can fly a straight line inside and explode in the middle (you are likely running several drones so each flys to a pattern hoping to get something useful inside). While flying to the middle they can do some image recognition, and if something looks like a high priority target they can target that, if not exploding in the middle will do something. The more powerful computer you put on (were powerful is often more expensive though not always) the better you can find targets

The important point in this is the drone will explode before it could reach anything not a target. It can sometimes find a better target than a pattern. In the ideal case you might fly it all the way, but if you lose radio over enemy territory anything the drone can find needs to die anyway so it may as well attempt to find something and kill it.

mike_hearn 5 days ago||||
They don't need processing power. The CV can be done at the operator site, allowing operators to handle more drones simultaneously.

It's useful for the fiber powered ones that can loiter indefinitely watching out for other drones and then go chase them.

msgodel 6 days ago|||
How much are the controllers they're using now? It's not like computers and cameras needed to do interesting CV are all that expensive.
pzo 6 days ago||
some mid-range iphones or android would even do the trick. especially iphones have tone of processing power nad strong NPU/GPU and lots of cameras, lidar, depth sensor, and plenty of other sensors. second hand phone 13 mini would do the trick and you can get it for less than $500
palata 6 days ago||
It's not as if you could glue an iPhone 13 mini at the front of the drone and call it a day, right?
Tepix 6 days ago||
We're rapidly heading towards the world that was warned about in the 2017 short Slaughterbots https://www.youtube.com/watch?v=9fa9lVwHHqg
xoatic 6 days ago||
International volunteer? I think mercenary sh*t would be more appropriate.
More comments...