Hacker News Reader

Comment by badfutureahead | Hacker News Reader

Today it's basically a week-long project to train a neural network to fly towards a jet engine in a 3D simulation. Then you just need to hook that to the drone control software.

No amount of jamming helps you here, this is radio control off, GPS off, entirely visual "guided missile" approach.

And this approach scales, with 20 drones, 2 drones assigned to each engine, you can engage 5 taking off planes in a row off a busy airport. And you can pick the direction which will make the planes crash in the nearby city.

With a "hire people to do stuff for bitcoin" project (there are several) you can prepare this in a bunch of countries (you just tell them you need some drones setup for an aerial shooting/light show or something).

The future is scary.

View on HN

"fly towards" doesn't scale too well with the huge speed difference. Your intercept cone is super narrow when jets travel 10x your quads speed. Then factor in altitude and this approach only works near airports, where defense is arguably easiest.

where this works really, really well is with boats. Iran did this with 15m RHIBs.

I work on these problems directly. Its scarey but mostly because we're not thinking about it, not because it's something we can't prevent. Don't get me started on all the truly horrifying things commercial drones are capable of even worse than airline terror.

With FlightRadar24 historical data you already know exactly where regular flights will be, so you just wait there. The only thing left is that the pilot will try to dodge you, but then you have automatic drone control versus human (with low visibility).

I agree with you that this is a high touch attack, a kind of "drone 9/11", versus other low hanging fruits like dropping drones on peoples head.

Humans aren't as regular as to allow this. The Mombasa attack failed because the pilot chose a different path that time, and that was an 'easy' attack (takeoff where the plane is slow, using military-grade AA missiles far quicker than any drone).

Drones have 10 to 30 minute loiter time, and they don't reach 10k feet. You're overestimating the ability of small craft to detect, track, and even reach the operational area of large passenger jets. These things are moving at hundred or more meters per second vs a drones ... 15? By the time a vision system had enough pixels on target to detect, it would have seconds to move into position in front of it. And how exactly is it supposed to know the jets velocity and heading?

Let alone it must hit the inside of the engine to do more than become a scratch on the paint.

Still, takeoff and landing maybe ... But still this kind of visual detection, tracking, and maneuvering is harder than you think.

Let’s not forget what’s possible 20 or 200 years ago. You can poison water supplies, you can stab people on the street, you can crate rail guns shooting out nails, you can drive giant semis into crowds of people, you can bomb or otherwise derail trains, etc etc. The ways to inflict carnage are great. But it doesn’t happen all the time, because most people aren’t seeking out murder on a repeated basis.

Yes we should find ways to defend against new kinds of attacks on common infrastructure that causes asymmetric levels of worry, but so recognize humanity has always had the potential to do damage.

The thing is: with most of what you're talking about: the level of risk taken was pretty high. With drones and other kinds of automated devices, the risk is going to drop near zero.

In other words, the real life will be as scary as the internet.

With this attitude, anything is a week long project. 1. You're making extremely optimistic assumptions about the results of your system. The translation between your simulation and reality may not be so smooth. Also jets are fast. 2. Lasers jam your system. 3. It's a difficult problem to crash a malfunctioning plane in a certain area when you are behind the sticks, let alone by doing it with a collision.

In the context of asymmetric warfare this might not be a significant issue. The offensive side need only be successful a fraction of the time, but the defensive side must be successful nearly all of the time.

Being successful is not binary. The offensive side needs to be significantly more successful in order to gain an upper hand. Otherwise it's defenders advantage every time.

And yet, the odd successful terrorist attack makes word headlines.

It's almost like there aren't actually vast numbers of motivated terrorists trying to kill us.

In Clausewitzian warfare this is true but not in asymmetric warfare. As a simplistic analogy, there's no need to score a knockout punch if you can trick your opponent into falling on their face.

Tbh, I think the only stable outcome is that which was during the cold war, MAD.

That or may be people can try diplomacy for once.

If you have money you can just launch 50 of them and hope 2 or 3 find their target. The marginal cost is relatively low.

So then the airport sends friendly drones to intercept the nasty drones. It's drones all the way down…

Exactly as Neal Stephenson predicted in Diamond Age.

Let's not forget brute force electronic means (disrupt drones' control circuitry) and mechanical means (obstacles, nets, dust, corrosives, artificial wind). We have many options to defend against rogue drones if motivated to do so.

Note that an airplane with a single engine out, even during the initial accent, typically does not crash into a city. The Hudson river landing occurred because all engine power was lost, not because of OEI (one engine inoperative).

Contained single-engine failure / blade failure and explosive charge inside engine are two different cases.

Modern turbofan engines are designed to contain a total blade failure. I'm not aware that they're designed to mitigate a charge detonated within compressor or turbine stages.

What's intriguing for me is that potentially unjammable machine vision drone missiles dramatically reduces the killchain. You don't need trained operators to loiter within radio range. And I think if anything it will heavily bias economics more towards attackers. It's already reducing relatively cheap hardware by removing training, operation costs and remote control hardware. Doesn't take much imagination to see how you can lock down a military airstrip with this tech or dramatically increase cost of defense. Albeit I am far from an expert in the topic.

I imagine the countermeasure will be cameras with computer vision controlling microwave lasers.

This is bit on a tangent but it’s bit amusing.

The term for that is maser. That term came before laser which was derived from it.

And now we have come the full circle and are calling masers microwave lasers instead of lasers ”maser but for visible light”

tercon — terrain contour matching does not need a camera. Earliest systems had a single radar sensor and an analogue feedback loop, and those were getting EEPs of 25 x 15m

And now we all have pocket sized supercomputers...

Reply on news.ycombinator.com