[aaronharnly]

This sentence was a bit cute: "Waymo has received our pilot permit allowing for commercial operations at San Francisco International Airport." Yeah, that kind of pilot.

I really had to read through it twice to make sure they were just talking about car taxis picking up travelers, rather than some kind of prototype pilotless commuter helicopter or something.

[danielvaughn]

That was my first interpretation, and I was very surprised and kind of afraid. Glad to know they aren't trying for autonomous flight yet.

[bdcravens]

I have zero expertise for my claim, but I feel like autonomous flight is easier than autonomous driving.

[jerf]

The hard part of automated driving is dealing with all the ground clutter that planes serenely fly over. If pedestrians could charge out in front of a 777 going 650 mph at 34,000 feet... well... we'd be living in pretty different world! And in that world, flying would be much more difficult. Not just for computers but for humans too.

Flying is obviously much harder than driving, but it's a sort of harder that is generally more amenable to automation, though I still think pilots are a good idea because when it goes wrong it goes wrong much worse.

[bluGill]

Flying is almost always easier than driving. landing is hard. Bad weather is hard. But just flying - human pilots have napped many times over the years and it only rarely is an issue. Airplanes with primitive autopilot are very good.

[Sohcahtoa82]

Yeah, a primitive autopilot in a plane just needs an altimeter and compass, but a AoA sensor, speedometer, fuel level sensor, and pitch sensor help to detect unsafe conditions like runaway pitch, stalling, overspeed, low fuel, etc. Each of those sensors is providing a simple 1-dimensional data point. Redundancy is relatively inexpensive.

Automatic lane keeping in a car requires cameras that software needs to then analyze to find the lines in the road in real time. But if you want a "set it and read a book for an hour", then you have to respond to other traffic. No longer just some simple PID controllers, the software now needs to plan and execute based on surrounding traffic.

[kersplody]

Yep. 0ft-1000ft AGL Takeoff, Climb, Approach, and Landing are the tough bits. The rest (Cruise) is very low demand and much easier than driving.

[toast0]

Taxiing is probably harder to automate than the rest. But you could have pilots on hand to taxi to the runway, and take a shuttle to the other end and hop on a just landed plane to taxi to the gate. Or you could use tugs for ground movement.

[Barbing]

>human pilots have napped many times over the

months?! :)

"The [German pilots'] union said it had carried out a survey of more than 900 pilots in recent weeks, which found that 93% of them admitted to napping during a flight in the past few months."

-The Guardian, "Almost all German pilots admit to napping during flights in union survey"; 2025-09-10

[bluGill]

Years as since humans have flown planes stable enough not to need constant attention. On a calm day you don't need autopilot, just set your trims correctly and some airplanes will hold course well enough for a short nap - though of course this is more likely to result in a crash (which likely has happened, though it is hard to guess why a plane crashed beyond pilot error)

[andrewstuart2]

Not to mention that almost all civilian planes in the US are required to broadcast a bunch of details that include their coordinates and altitude on a public channel (ADS-B). It's the kind of automated collision avoidance input that you'd probably dream of as a self-driving system engineer. Basically the only thing you'd need to avoid via more complex systems is the odd military traffic, small craft at low altitudes, and birds.

[0_____0]

In the abstract yes but in practice the economic (ratio of cost of pilot to pax miles) and safety context of aviation mean fully autonomous flying has to be extremely robust before it has actual utility in industry.

[rkomorn]

In practice, you're also currently very reliant on infrastructure that is definitely not as solid as you want (eg: ILS and GPS can be interfered with quite nastily).

ILS being under maintenance and unavailable for certain runways is also far from unusual.

[notahacker]

Commercial pilots are also extremely good at dealing with edge cases you wouldn't design an autonomous system for no matter how solid the infrastructure, like deciding the Hudson river is a good place to ditch

And their cost relative to other operational costs is so low there isn't even any pushback on regulations regarding there being two of them.

[rkomorn]

Pilot cost isn't low. The airline industry is very much looking for ways to reduce crews, whether that's going to single pilot operations (long term) or reduced crew operations (short term).

RCO is very much "pushback on regulations".

[dcrazy]

On the happy path, yes. Though I don’t think takeoff is automated yet.

Currently we rely very much on the problem solving abilities of human pilots to deal with troublesome situations. Autopilot will disengage in many scenarios.

[seanmcdirmid]

I'm pretty sure drones can already take off on their own. Taking off is a lot easier than landing, and planes have auto-landing tech already.

[nradov]

Drones (both autonomous and remote piloted) have much higher mishap rates than crewed aircraft. Taking off is "easy" until something goes wrong, like a mechanical failure or runway incursion. It's impossible to anticipate and explicitly code for every possible failure mode, so developing autonomous flight control systems that would be safe enough for commercial passenger flights is extremely challenging.

Category IIIC ILS (full auto-land) does exist but requires special equipment for both the aircraft and airport. Human pilots have to actively monitor the system and take back control if anything goes wrong (which does happen).

Garmin also has the Autonomí auto-land system for certain general aviation aircraft which can attempt to land at the closest suitable airport. But this is only used for single pilot operation in case the pilot becomes incapacitated. It isn't suitable for regular flights.

[lolc]

Consider that drones may fail more because failure is an acceptable outcome for drones.

[dcrazy]

Takeoff at a commercial airport is a very challenging and potentially dangerous situation. There’s way more margin to abort a landing than a takeoff.

[prmoustache]

OTOH takeoff and landing could in theory be operated by people on the ground, flying simulator style.

I still believe that having an actual pilot inside the plane that care for his own life is not a bad idea vs someone remote feeling a bit disconnected with the reality of a crash.

[csours]

It's kind of funny how you can both be right.

[glitchc]

Drones crash on takeoff all the time. Worth noting that drones are more than just quadcopters and serious drones are often winged aircraft.

[lawlessone]

It's the failed takeoffs that lead more often to jets leaving the run way and crashing into buildings or trees.

[hoosieree]

0d (parked) - null program easy

1d (train) - easy. just one lever

2d (car) - hard. super hard. why is it so crowded? who thought this was a good idea? you let teenagers do this?

2.5d (plane at takeoff or landing) - almost as hard as car. fewer pedestrians.

3d (plane flying) - easy even with all those extra levers

[SR2Z]

I'm not actually sure how hard landing is. Most airports that support autonomous landings do it by having ILS antennae that guide the airplane to within tens of feet of the runway, at which point the airplane switches to radar for altitude.

Automatic landings started in 1964. I think that it seems hard mostly because of how tightly regulated aviation is - modern technology could probably make things a lot better if people were more receptive to the idea of heavy automated aircraft over populated areas.

[adgjlsfhk1]

landing is easy. the hard part is landing with 20mph cross winds and one engine out (or other mechanical failures). we've had auto-land that is 99% reliable for a while now, but you need to get to 6 9s before you have a system safe enough to replace pilots

[SR2Z]

I think that as long as the autopilot is able to fly in a crosswind or with an engine failure, it can probably land with one. Autopilots are already able to do these things.

I doubt anyone has tested this in depth, but I'm not sure there are too many configurations of airplane these days where a human can safely land it and a computer can't. Maybe if a big chunk of wing or control surfaces were totally gone, but even a human pilot isn't getting 99% reliability in a situation like that.

In any case, I don't think that the first candidates for automation are gonna be passenger flights. It will probably be small cargo planes first - Cessna Caravans and other turboprop aircraft where the cost of paying pilots is roughly similar to the price of fuel.

[anticensor]

1d, variant (tram) - hard, who thought it was a good idea to send rail-bound vehicles and steerable vehicles down the same road?

3d, variant (orbital) - super hard, so hard that trajectory pre-calculations has to be performed

[themafia]

When everything is working correctly, no other pilots have emergencies, and no temporary restrictions are in place, and there are no clouds in the sky. Then yes, it /could/ be easier, but almost always it never actually is.

There's a reason the majority of accidents occur during take off and landing.

Spend some time listening: https://www.liveatc.net/search/?icao=ksfo

[tim333]

It depends a bit on your safety standards. There are already autonomous flying things delivering blood and blowing up oil depots where it doesn't matter so much if stuff goes wrong, but to be an airline pilot you have to know how to deal with a huge range of emergencies and systems packing up.

With a car if the engine fails you just pull over. With an airliner it's not so simple. As a result the training for a pilot is much longer than for a bus driver say.

[ugh123]

Navigation might be easier. The battery and safety tech isn't there yet to make it practical.

[njarboe]

Not a pilot myself but it seems that a large part of the danger with flying is that when something goes wrong you are much more likely to have a high speed crash. Cars don't even usually travel at speeds that planes crash at.

[tialaramex]

When shit goes wrong for a car (such as Waymo) you just stop. Now, that's not trivial, but it's also not very difficult, I expect most of it can happen even if the Waymo hardware itself were suddenly destroyed, rolling along is the hard bit so not doing that isn't too difficult. Everybody aboard can just leave when it stops moving.

In contrast when shit goes wrong for a plane we've got a big problem. Just stopping will definitely kill everybody, even from a modest altitude at a very low speed suddenly plummeting to the ground will straight up kill you. So, we want to land, albeit maybe we have to "crash land" destroying the vehicle to perhaps save its occupants.

You can buy (and indeed to some extent you can even retro-fit) emergency auto-land for small planes. Once engaged, or if set to do so automatically upon pilot failure the plane will figure out where it is (using GPS), pick the emergency radio frequency and announce the problem and its intended solution (I am a machine. My human pilot is incapacitated. I intend to fly to X location and land there. I am not listening to you and cannot understand you) and then it will fly to a chosen place and attempt to properly land the aeroplane, broadcast on radio that this airfield is now closed (this aeroplane is parked on the landing strip so you can't use it!) and then switch off.

Maybe the pilot is still alive and human medics can rush them to hospital. Otherwise maybe there are passengers who have been saved. In any case at least the aeroplane is now on the ground where humans can easily take over e.g. moving the plane so the airfield can re-open.

[kbenson]

My layman's understanding is that we've been doing it already for decades with expert system "AI", so likely much easier than navigating streets with other people.

[anonymars]

In a pinch, a car can just put on its hazards and pull over

[dcrazy]

That “just” is doing some heavy lifting! The car still has to deal with all the normal hazards of the road while pulling over, plus the hazards it is itself creating by acting abnormally.

[anonymars]

Well if we're being picky, technically the car itself doesn't have to deal with the hazards it has created, rather everyone else does.

The point is you can't just "stop" a plane and wait for someone to figure things out (https://support.google.com/waymo/answer/9449023?hl=en). Whatever the difficulties in dealing with an abnormal situation in a car, it is strictly much more difficult to deal with them in a vehicle constantly fighting the homicidal urge to fall out of the sky.

[Dylan16807]

They each have their own unique issues. Being in a pinch is not universally harder for a plane.

Also constant urge to fall out of the sky is a helicopter. A plane generally wants to glide.

[e_y_]

Could also be a big challenge if you have dozens or hundreds of autonomous cars in the area that need manual intervention to get them out (plus the people who get stuck there)

[anonymars]

Is that situation somehow less difficult with aircraft?

[efavdb]

Don’t have a ref but heard that it’s been safe for quite a while but they keep the pilots around due to consumer fear rather than actual improved performance. Curious if anyone can confirm.

[csours]

If you can design the product and environment to fit automation, then automation can be quick and effective.

The less you can change about the product and environment, then automation run slower and less effectively.

Air liner operations could be automated, but the minimum equipment list would be more stringent, the destination airport would not be able to take any equipment out of service for maintenance, visibility minimums would increase, takeoff and landing operations would require more slack time.

Besides all of that, the owner of the airplane would still want to have some crew on board.

In short, it's not worth it yet.

===

There is also the paradox of automation: Automation generally makes the hard parts harder and the easy parts easier.

[rkomorn]

The current goal of autonomy for airliners is single-pilot operation more than full autonomy.

It's very cool stuff, technology wise, with potentially significant redesigns of cockpits, etc.

But the main thing is the plane basically needs to be able to operate just about entirely autonomously (especially during critical flight phases) in case the pilot is incapacitated.

In theory, once SPO is solved, autonomy is almost solved.

[nradov]

I'm skeptical that SPO will be allowed for commercial airliners in our lifetimes. Pilot workloads are fairly low during most routine flights. But when an emergency occurs then the workload suddenly gets extremely high, to the extent that even two pilots are sometimes overwhelmed. This isn't a problem that current automation technology can solve. There are an infinite number of possible emergency scenarios and engineers can't possibly code for and test every one.

Cargo flights over oceans and (mostly) unpopulated areas might be a valid use case for SPO. Cargo pilots have always been considered somewhat expendable.

[rkomorn]

No. Airliners can't even take off on their own yet, and are only allowed to auto-land with zero visibility at a few dozen airports when the pilots, plane, and runway are all current/recently checked.

Look up the Airbus ATTOL project's first automated takeoff a few years ago.

Also, there's virtually no automation when it comes to interacting with ATC.

[SoftTalker]

An airplane will take off when it is properly configured and it hits a certain speed. It's simple aerodynamics/physics. Pilots are there to react to failures and unexpected events.

[3ple_alpha]

There's a bit more to it since you do need to do last bit of configuration (pull up the nose) just as you hit the target speed. But yeah, automatic take-off is quite a bit easier than automatic rejection of take-off.

[rkomorn]

Sure. It'll also land if you don't care about anyone surviving.

[glitchc]

And the air is within acceptable temperature and pressure ranges. I assume configuration takes weight into account as well.

[johnisom2001]

> Also, there's virtually no automation when it comes to interacting with ATC.

Check out the Cirrus Autoland feature in their aircraft. They are all small personal aircraft, but the tech is pretty cool. Will talk to ATC and fully auto-land for you in the event of an emergency where the pilot is incapacitated.

[kotaKat]

"Talking to ATC" is a bit of a huge ask. The system basically just hops on 121.5 (and maybe the nearest/local unicom/tower frequency) and start an automated callout with its intentions that it will be doing. It operates on the assumption that all other airspace users will hear the radio calls and stay clear of the emergency aircraft.

[rkomorn]

I'm aware of it, though I've never flown a Cirrus. But AFAIK, it announces what it's doing. It's not communicating.

[KeplerBoy]

What would ever make you think that?

In an automotive setting you can almost always safely decelerate to a full-stop, put on hazards and call it a fail. Good luck trying that in an aircraft over urban areas.

[snickerdoodle14]

I also feel like the demand is way, way lower. A pilot can't be that large a % of the cost of a flight. Maybe if we lived in the jetsons era.

[seanmcdirmid]

The problem is actually safety. As automated systems get better, the pilot is left with not much to do, and has to maintain vigilance while being really really bored. It is almost better to have fewer automated systems and give the pilot more things to do during the flight so it is easier to keep them paying attention, or all automated with no human pilot to mess things up.

[rkomorn]

Depends on the size of the plane, really. One of the reasons a few companies were investing in fully autonomous air taxis is because the math on a small piloted aircraft wasn't realistic for a low enough price point to be competitive.

[amelius]

"Autopilot" already exists when it comes to flying.

[bdcravens]

Sure but it's not autonomous in the sense of Waymo (ie, driverless)

[ckastner]

Landing can be: https://en.m.wikipedia.org/wiki/Autoland

In fact, it's pretty routine. Don't have the source at hand, but somewhere around 1% of all landings (at airports with ILS) are autolands.

I think it was Boeing that even requires at least 1 autoland per plane every 30 days or so.

You can find videos of this on YouTube. Completely hands-off.

[themafia]

Most carriers have a rule that on clear days you always hand fly the landing.

This is a competence you do not want to lose.

It's also the case that you can have a whole approach setup in your flight computer and at the last minute the controller gives you a runway change. You could drop your head down and start typing a bunch info the FMC but you're generally better off just disabling auto pilot and manually making the adjustment.

[amelius]

I'm curious, what is harder to implement: autoland for airplanes, or autoland for rockets (spaceX)?

[ckastner]

I don't know if these are comparable.

But two interesting data points from the Wikipedia article I linked are that the first aircraft certification for ILS Cat III was in 1968, and Cat IIIB in 1975.

And IIRC by the 1980s, autoland was already a pretty common feature.

[danielvaughn]

Yes but it should have been obvious that in the context of Waymo + SFO, the implication was autonomous flying of commercial airlines.

[seanmcdirmid]

Yes, but autopilot usually just keeps the plane flying in a straight line at some specified altitude, which have been around since 1912. It isn't full self-flying (although we definitely have drones that can fly themselves already, so that tech already exists).

[dawnerd]

That's an oversimplification of autopilot systems. They can follow flight routes, avoid traffic (TCAS), even auto land to name a few.

[seanmcdirmid]

Auto-landers are not simply classified with autopilots. An autoland system is an advanced function that is part of a modern aircraft's overall autopilot capabilities. A basic autopilot can control an aircraft's attitude and heading, but an autoland system can automatically execute the full landing procedure.

[Fomite]

Mine as well, and I was crossing SFO off the list of airports I'd connect through.

[blackkettle]

Cool, I wonder if this means they will finally start letting foreign visitors also use the app. I'm an American living abroad now for many years, and I was initially super excited to try Waymo in LA and SF this summer when I visited with my family. Unfortunately they only make the iPhone app available via the US app store, and while I actually have a US credit card that I could have in theory used to make the switch, Apple makes it an absurd pain to change your region as they require you to both a) cancel any existing subscription AND b) wait until they all expire. Most tourists have it worse as they have no option to even switch in theory.

[nicoburns]

Huh, as a Brit, I was able to use Waymo just fine on this summer on my Android device.

[dlcarrier]

See also: garden-path sentence (https://en.wikipedia.org/wiki/Garden-path_sentence)

[seany]

This is actually why you have "Naval aviators". To maritime people a pilot means something else. https://en.wikipedia.org/wiki/Maritime_pilot

[loeg]

The flying kind is a license, not a permit.

[simonbw]

There are driver's licenses and learner's permits. This could be the flying equivalent.

[ricksunny]

lol deniable demand-gauging :)

[tdeck]

Honestly I think the title should be edited. The first time I scrolled past it I had the same obvious interpretation.