[fallingknife]

Setting a protocol to handle air traffic control and collision prevention in airspace around airports is a 100% automatable problem. You don't even need a centralized control system. This can be handled entirely with software running on each plane. Same way a flock of birds can fly and never collide with each other.

[viraptor]

Unfortunately that's not how things work in practice https://en.wikipedia.org/wiki/Split-brain_(computing) If Jepsen fails every database coming from single source, imagine the chaos of synchronising a P2P of various clients of various versions over a very noisy link. We can't even achieve that with home automation meshes that send maybe 3 types of messages!

Also you need to handle planes without computers - you can land a personal plane at almost any airport. (With lots of caveats but still) Also you need to handle planes with failing automation. Also you really want to know the situation on the runways, so there's really no need to remove the single source of truth here.

[fallingknife]

We operate cars on the road with not only no centralized system, but also minimally defined and enforced protocol, and yet Waymo has achieved a near zero collision rate inside a swarm of cars that are not running equivalent software. And this is in a situation where cars are only a few feet from each other while operating at top speed. So you can come up with a million objections but they are all solvable. As for automation failure, the rate of that can be easily made lower than the rate of human failure, which currently is fatal to a plane.

[const_cast]

Waymo is significantly more dangerous than air travel to a degree that the comparison is actively offensive to everyone currently alive on Earth.

We can't even get cars, working in more-or-less two dimensions, to go without constantly running into each other and being one of the major causes of death in human civilizations. Waymo "solved" that problem in about, oh let me see here... yeah 0.0001% of cases. So, we're almost there! That's, like a couple cities out of all cities on Earth.

Yeah that's bad. Really, really, really bad. Like so bad it's not even worth talking about and comparing to air travel.

[viraptor]

That's a completely different system. In a car everything still resolves in seconds and needs minimal planning -corrections can happen in real time. That's completely different from "there's a queue of planes coming in and you have to manage the throughput available on the ground". You can't just hit emergency breaks on everything and slowly resolve the situation. These things are not comparable.

[lurking_swe]

99.9% success rate is probably not good enough when you consider that the vehicle in question costs $200 million, and has 150 - 350 humans on board. It’s not a “whoopsie” like with cars where the damage is $50k at most (waymo vehicle) and maybe 2-3 people injured or dead. Also dead pilots who are already in short supply.

Also cars only care about what’s going on in a horizontal plane. 3D space is clearly more complex and probably requires more computational power i’m sure. Consider that boeing couldn’t even correctly write software to keep their 737 MAX planes from crashing into the ground (MCAS software issue). Something that simple was too hard for them. Speaks volumes doesn’t it?

Not saying it’s impossible. Just saying that this clearly isn’t a “just do it” problem. Waymo’s been working on their software for how many years now? and they still have minor crashes lol. Not an easy problem at all.

Also consider that airlines usually outsource their software development so they don’t have cutting edge expertise in house.

[lurking_swe]

So what happens when a plane has a critical failure (related to this P2P communication), how does it land? How would other planes nearby magically know what the plane that is in distress is going to do? It’s basically an unpredictable peer in the network.

YOLO i guess? :)

Hypothetically the nearby planes can detect that unresponsive plane on radar or other sensors, and try to react together as an intelligent swarm, to avoid it and let that plane land manually. But it’s not so simple. Planes are not loaded with full fuel tanks, only a bit extra. Some planes may have already underwent a go-around if the airport is busy. So it’s not just “land without crashing”, it’s also a prioritization issue.

IMO we certainly need humans in the loop, in a centralized fashion, to “orchestrate” a manual emergency landing if there is some critical cascading failure or bug in the software. I agree that in the happy path (99% of cases) it’s possible to automate it all. In theory.

Things get more complicated when you consider that small planes (flown for hobby, flight school, etc) have waaaaay less tech. That can’t work in some peer to peer fashion without a major upgrade to all those planes too. And the owners of those planes are not corporations making billions.

[trehalose]

What happens when an airplane's pilots have to radio ATC to request an emergency landing, and the planes' sensors have failed so it can't safety land itself?

[fallingknife]

If the plane is able to communicate with ATC it is able to broadcast to the other planes that it is on an emergency landing, so this is not an issue. Even if it has lost all communication this is still not an issue because all the sensors of the other planes can see it. So if the disabled plane immediately goes for an emergency landing, all other planes in the area are able to see its position and that it is not responding to pings, and therefore set safe courses that avoid it. This really isn't a very difficult problem.