Hacker News Reader

Comment by greedo | Hacker News Reader

Drones are nice when you have a permissive environment where the locals don't have a decent air defense system. But otherwise, the current models we have are relatively easy to shoot down. They lack speed, maneuverability, and their operators lack the situational awareness that a pilot would bring to the table.

There are better drones being developed, but they will be expensive and will depend extensively on sat links for communicating between the drone and base station. This will be problematic against a peer foe who can jam radio transmissions.

View on HN

I wouldn't trust an autonomous drone in many environments, but air warfare is pretty friendly to AI. And while humans are limited to 9 Gs of acceleration drone aircraft that can turn much more rapidly can and have been built. In the future I expect we'll see gaggles of drone fighters overseen but not piloted from a AWACs far back but still within visual range.

Those drones are still flown by human pilots, remotely. I wonder what the maneuverability limit is with remote piloting. At some point the drone would be able to turn so fast that the human couldn't process the rapidly changing visuals. Maybe the drone pilots will need to be given a simulated "third person view" like a video game in order to keep up.

It's an UX problem. The pilots might need to be given more indirect, goal-oriented interface. Instead of first-person view and direct flight control, they would mark waypoints on situation map and select maneuvering policies, akin to some RTS games.

I think modern wars will be fought by tousands of pro-gamers using something like 3d rts ui.

For example if we have 50 drones doing some task it's stupid to hire 50 people to manage them like pilots. Much better to hire 5 people and give them better ui (construct 360 degree view from all the drones feeds localy and send only the aggregated information.

With current technology, humans can fly in worse weather than drones.

And still human errors seems to be directly involved in the worse crashes (see the AF 2009 flight to Brazil: https://en.wikipedia.org/wiki/Air_France_Flight_447#Human_fa...)

AI isn't anywhere near being able to discriminate against an aircraft that is not exhibiting "aggressive" behavior. How will it work when it needs to visually ID an opponent, or when the opponent is flying aircraft that might be allied? Say our opponent is flying an F-16, or F-15? Lots of those out there, and in militaries that easily could become hostile. IFF is not always reliable, (hence the viz ID requirements).

And creating an autonomous UCAS capable of conducting air to air combat isn't going to be cheap at all. The reason we like drones now is because they have the performance characteristics of a Cessna, and they don't put pilots at risk. To compete in A2A combat, they'll need performance (and avionics) that exceed what our best fighters currently have. That will be buku $$...

It's enough for the drones to be semi-autonomous. The controllers mark detected airplanes as friend/foe manually, and on-board AI does the maneuvering and shooting.

And what happens when the controllers are jammed? Or when they can't detect whether the bogey is friend or foe? The Air Force lost an RQ-170 to the Iranian Air Force because it relied on commlinks that are susceptible to jamming. If the comm links are vulnerable, how will the "controllers mark" the aircraft?

We don't currently have a problem with enemies jamming comms between AWACs and the F-15s or F-22s they direct. The RQ-170 was using a satellite uplink which is a lot easier to mess with than an airplane closer by. Using jamming against combat aircraft is also problematic because jamming platforms are impossible to hide and very vulnerable.

Correct. And that's because we're fighting an opponent that has trouble with high tech, and has limited resources. If we end up in conflict with Russia, or China or another opponent with better capabilities, your links are going to be severely degraded.

Out of interest once they reach a certain level of basic ability wouldn't it be easier to have them in "packs" rather than trying to basically replicate everything a human pilot can do in a single drone?

If you have a "pack" of drones big enough that a single enemy combat aircraft can't take them all down at once then if an enemy aircraft were to fire on a drone in the "pack" it could automatically be designated an enemy and be hunted by the remainder?

> AI isn't anywhere near being able to discriminate against an aircraft that is not exhibiting "aggressive" behavior. How will it work when it needs to visually ID an opponent, or when the opponent is flying aircraft that might be allied?

What will you do when the enemy sends up unmanned autonomous aircraft that ignore these niceties and shoot down all your manned fighters?

I'm not dismissing the legitimate problems you raise, but we need to prepare for war, which almost by definition is the absence of rules, such as distinguishing military from civilian targets.

EDIT: Also of interest, if an autonomous fighter shoots down a civilian aircraft, is it a war crime? Mistakes happen in war that aren't crimes. And if it is, who goes to jail?

If the enemey's autonomous aircraft can't discriminate between friend and foe effectively, then when you launch a swarm of them, they'll all fratricide each other. It's a fundamental requirement that they be able to differentiate between friend and foe.

> If the enemey's autonomous aircraft can't discriminate between friend and foe effectively, then when you launch a swarm of them, they'll all fratricide each other. It's a fundamental requirement that they be able to differentiate between friend and foe.

I would guess that identifying your own teammates is much easier than distinguishing strangers who are combatant from non-combatant, and among combatants distinguishing allies from enemies.

Also, suicidal autonomous aircraft are common, in the form of missiles. If these more intelligent missiles kill the enemy, it doesn't matter if they kill each other too (beyond the financial loss). Maybe you give each a 'kill area', in which they are instructed to kill everything, and spread them out enough that they don't overlap too much.

>buku

Completely off topic, but did you mean the French beaucoup to imply the phrase "Lots of money" here?

To a degree he did. http://www.urbandictionary.com/define.php?term=buku

Oh cool. Didn't realize it was considered a word by itself. I guess it can be considered slang since it's on urban dictionary.

Yeah, maybe it isn't easy to make the AI distinguish "aggressive behaviour", or even do good visual ID. But how's this for a use case: Simply declare a restricted airspace, where every plane needs to have transponder replying correctly to some challenge-response authentication protocol. Distribute transponders to allies. Tell about zone in the evening news, so that everyone knows. Then program drones to shoot down anything that doesn't respond correctly. I'm pretty sure that it's a lot cheaper to design and build drone with same performance characteristics as top A2A fighter, there's so much that you don't have to worry about: human rating everything (you don't have to make so damn sure that everything works without the hitch first time), all the control systems that you have to build the pilot an interface to (radar, etc.), the cocpit.. I'm sure those will easily pay for developing a competitive fighting-AI.

That's why the US is developing tracked laser communication systems. They are effectively impossible to jam.

Laser is line of sight, limiting your range. And nature has a great way of jamming lasers, called clouds. If you were able to get a decently powered laser in a satellite, it wouldn't be able to control too many aircraft unless it looked like a porcupine. And then it would have to deal with anti-satellite defenses that both China and Russia have spent significant sums on.

Not sure why this was downvoted.

Unless you happen to have an autonomous reusable spacecraft with significant cross range ability and enough fuel to continuously change orbits.

> Laser is line of sight, limiting your range.

Yes, with current equipment it limits your range to somewhere beyond the orbit of the moon. Yes, we can now keep a laser and a receiver tracked on each other from 300000km away.

> And nature has a great way of jamming lasers, called clouds.

There are wavelengths that are not hampered much by clouds.

> If you were able to get a decently powered laser in a satellite

We can.

> it wouldn't be able to control too many aircraft unless it looked like a porcupine.

... but this is a problem, you'd basically be limited in total bandwidth by count of satellites. Meaning you either need to launch a lot of them, or reduce bandwidth per drone.

Being able to ping the moon with a laser doesn't help with terrestrial systems. Just as radar has a horizon limit, so do laser systems. So if you put this on a controller version of an AWACS, you have to deal with attenuation from clouds/weather, and I think keeping a lock on a maneuvering UCAS might be a challenge.

I'm not talking about pinging the moon, I'm talking about maintaining a link with a maneuvering spacecraft in moon orbit.

So this spacecraft you're linking to makes radical 8g maneuvers, changes direction at a 30s/ roll rate, and can change altitude at will?

Reply on news.ycombinator.com