Hacker News Reader

Comment by swamp40 | Hacker News Reader

I don't see fly-by-fiber getting very popular. The only goal is anti-jamming. But if the drone was fully autonomous, there would be no signals to jam. And the fully autonomous drones are coming fast. Meta's SAM 2 can follow pretty much any object anywhere, and people are beginning to get it running on the NVIDIA Jetson Orin. That's 75% of the work. The other 25% is autopilot and a way to disable it if it flies back over friendly territory. 6 months I'd say, for amateurs starting now. I'm sure some companies already have prototypes working.

View on HN

There are multiple roles for a small airborne asset. One of them is ISR (video feed), used for targeting artillery or movement tracking and such. Jamming can block the down stream feed and defeat that support role.

Anti-jamming is the intermediate goal. Fly by fiber allows the pilot to retain control of the drone while over the battlefield.

With a completely autonomous drone in contested space (presumably with EW around), there's no way for a pilot to tell the drone "fly a little to the left to see what's behind that tree." Each drone has very specific use cases.

That just moves the arms race from anti-jamming to anti-ML targeting, doesn't it?

Litter your trenches in inflatable mannequins (anti-CV), put all your soldiers in ghillie suits (anti-CV), have cheap fireworks ready to spread chaff (anti-radar) and flares (anti-IR).

The obvious solution is to target civilians with autonomous drones. Not much different to sanctions in terms of civilian casualties, but much more effective at stopping the enemy infrastructure. Since sanctions mostly kill children and the elderly.

Ah, war crimes.

At that point, isn't it easier to just shell those civilians if they are in drone range, or cluster/fire bomb them if you need to fly in an aircraft to deliver the drones anyway?

Terrorism by non-state actors is a danger, of course, but guys with AKs and pipe bombs storming a football stadium seems easier to pull off and yet hasn't happend so far anyway.

What's the point of doing that? What kind of mission objective does that achieve? If you want to scare civilians away or deport them, you don't need to kill them at all, you just need to make them run for their lives.

The same reason you starve hundreds of thousands of children. To spread democracy of course.

Any reasonable autonomous control system is going to have sensor fusion to combine different wavelengths of sensor data together. Kalman Filters are a thing, even outside of military applications. You're going to need countermeasures that are anti-CV, anti-radar, and anti-IR measures in one source to confuse a well-made autonomous drone.

Even then, if they just put a bullet into anything that has the approximate optical & heat signature of a human, it'll work fine. Who cares if you blow up a few inflatable mannequins and flares if you also get all the soldiers?

This is very clearly divorced from the realities of warfare.

* A Kalman filter is an algorithm, not a magical wand. How is that algorithm implemented? What are the parameters? How are they tuned? How is the state, and uncertainty thereof, modeled? How accurate is that model in the field? Details and implementations matter. Finding answers that work reliably in the field, even for a limited set of circumstances, takes huge amounts of time, money and talent, and involves a continuous process of trial and error.

* Drones have a limited payload capacity, which means a limited amount of ammo to burn on false positives, and a limited amount of smarts and sensors with which to process their environment.

* Many attacks rely on the element of surprise to catch the enemy unaware, or leave them with little time to plan - shooting fake targets alerts others to your presence and can quickly give away your position, providing time to take cover and/or stage a counterattack.

* The people developing the smarts for these drones are a limited resource that need to be found, hired, and paid, and they have to play a cat-and-mouse game to maintain the robustness of the targeting system.

* Asymmetric warfare is a time-honored guerilla favorite. It doesn't matter how fancy and sophisticated your drone's targeting system is if rendering it useless is cheaper, and wasting ammo on non-targets can very much tip the scales here.

That system is already orders of magnitude more complex than a small autonomous suicide done with a cheap camera, and probably not within reach even for the big players for at least a couple of years. At least of you want to do that with a solid, reliable targeting system and have at least some friendly fire prevention in place.

And more importantly, that's what I mean with arms race. Now the mannequins in the trenches get a little heater element to keep them at body temperature, and they get internal water bladders with the appropriate radar cross section, ect. Easily doable today, and orders of magnitude cheaper than a sensor fusion drone.

Add counter drones and point defense cannons, and the attack drones need rocket motors for final approach, which makes CV another order of magnitude more difficult, ect

Autonomous can still be jammed. They can and probably will still use gps, visual info, sonar, sound, magnetic fields, or pressure. All which can be jammed from the outside. And this is already happening.

ISR is probably more important than hunter-killer drones, at least right now

Any fully autonomous drone that can be produced in the near future will be easily foolable. And if it has a remote fail-safe deactivation system the enemy can figure it out and deactivate it too. So you solved absolutely nothing of what a wire solves. And made the same pitch we have for every ML product right now: we're X% there, next year we're there. When in fact next year we're still going to be next year away from there.

There's a simple way to build a failsafe that the enemy can't figure out. The basic rule of cryptography--assume your enemy knows everything but the key.

When the launcher and the drone connect they make up a completely random key. It is known only to the launcher and the drone. The drone will self destruct if it receives said key. The enemy could jam the destruct but they couldn't trigger it. Note that no encryption is even needed.

One time pads are inherently unbreakable crypto other than by compromising the pad. And in this case the secret is known only to the drone and controller.

But if the destruct is jammable it's not a real fail safe.

One goal could be really high bandwidth. Wonder if you could power it somehow too (or would that need copper or other conductive metal?)

fly-by-fiber used to be very popular. And still is. The systems Russia is getting from North Korea is fly-by-fiber. Though I agree that autonomous drones is probably where we are heading. Ukraina has already shown swarms of autonomous drones working together.

You even have different types of drones providing different function. A mama drone carrying smaller drones to the place of operation for example.

Maybe look at real-world use of drones right now in Ukraine? Where it's literally life and death.

Which is humorous to me as consumer drones get these sorts of features, meanwhile the military industrial complex charges 100-1000x the price.

"But they have to test to guarantee safety"

Nobody cares, there's a reason Ukraine is using cheap consumer/commercial drones, because they're effective as hell, and you get much more bang for buck without most of your money going into the back pockets of defense company executives.

If national defense is sooooo important, why aren't defense companies publicly owned?

Reply on news.ycombinator.com