We've had truly autonomous long-range flight since 1998 with the RQ-4A Global Hawk https://en.wikipedia.org/wiki/Northrop_Grumman_RQ-4_Global_H...

A couple highlights from the Records section:

> On 24 April 2001, a Global Hawk flew non-stop from Edwards AFB to RAAF Base Edinburgh in Australia, making history by being the first pilotless aircraft to cross the Pacific Ocean. The flight took 22 hours, and set a world record for absolute distance flown by a UAV, 13,219.86 kilometers (8,214.44 mi).[97]

> On 22 March 2008, a Global Hawk set the endurance record for full-scale, operational uncrewed aircraft UAVs by flying for 33.1 hours at altitudes up to 60,000 feet over Edwards AFB.[98]

Unlike Predator/Reaper/etc. drones, the Global Hawk is autonomous, there is no pilot sitting in a faux-cockpit stateside.

Do you know how these drones communicate with the ATC to state their intentions? Is there a flight plan made ahead and plan that they don't deviate from? I could not find any information online.

The operator is communicating via VHF with ATC. I’m not sure on exactly how the transmissions are handled (via aircraft, or remote radio station?) but the operator is in communication with ATC. If you can find a live ATC feed of Syracuse tower they’re frequently doing touch and go’s there.

One thing to keep in mind is that any type of auto-fly is dependent on a LOT of automation systems on the plane itself being functional. For example, instruments, sensors, autopilot, auto throttle.

To be considered safe, an auto-fly needs to be able to gracefully degrade as the subsystems in which it depends start to fail. Human pilots are trained to handle this by compensating for system failures in a variety of ways, including compensation (eg in a split flap scenario) and disabling (turning off a misbehaving autopilot). Ultimately, a human pilot can always turn off all the automation and fly the plane by looking out the window (or, in poor conditions, using gyro powered “steam” gauges).

In the case that the pilot is disabled, then auto-land is a great option, even though it’s reliant on so many subsystems. But for an always-on auto-fly, the requirement to gracefully degrade, and degrade, and degrade, and still operate at the level of a human pilot using the “Mark 1 eyeball” is a tough one which hasn’t yet been met (as far as I know).

There’s mention in this discussion of fully automated drones. But, those are basically expendable if there’s a serious system failure, since there are no souls on board.

Gyro powered steam gauges are all but gone from aviation. The GA fleet is even shedding them at a promising rate. MEMS based independent systems with backup power are solid state and far more reliable than a janky vacuum pump driven by the engine running a couple clockwork instruments.

Sorry to be a pedant but since it took me a second to get it at first: "all but gone" implies that they're definitely still here, I think something like "almost gone" or "almost out of use" would be the right phrasing.

all but is an expression meaning almost. See e.g. https://grammarist.com/usage/all-but/. It's one of the most confusing English expressions I've encountered.

'all but' is an idiomatic synonym for 'almost'.

https://idioms.thefreedictionary.com/all+but

I don't see how that's relevant. _They can and will still fail_

What happens then they do?

You have more than one independent system. A typical glass panel install will have a primary ADAHRS (air data attitude and heading reference system which contains a MEMS magnetometer, gyro, accelerator, pitot static pressure transducers, and processor) and a secondary ADAHRS both connected to the primary flight displays. Should one fail, you have a fallback for your PFD. As a backup, were you would normally have a vacuum gyro, you have a small glass panel (the garmin offering is the G5) which contains it's own built in ADAHRS and battery and will continue running even with the entire aircraft electrical system disabled.

Most of the GA fleet is still vacuum powered for backup.

Why do people keep parroting this? I know of exactly zero aircraft currently produced that are retaining the vacuum system and I've seen very few retrofits that retain the vacuum on the engine. The aircraft in operation with vacuum systems use them as primary systems but are moving away from it with each annual.

>Because it would be awesome to charter small flights from an app to jump to a nearby city.

I doubt the cost of a pilot's time alone is a big enough factor to meaningfully change the number of people with access to chartered flights.

The other costs (hours of plane time, fuel, ramp fee, parking fees, maintenance) probably dwarf pilot time as a percentage of total cost. And you're either pushing the absolute unreliability of aviation [0] to the customers who are now trapped in a different city, or they're paying even more to the app to have backup plans ready to roll.

[0] Commercial aviation deals at a scale that can absorb flights that don't keep their plan. You go to the counter, get rebooked, and angrily eat some shitty fast food. When something goes wrong on a private flight, you're pacing around on the tarmac at 1 am, 3 states away from home and have to be at work in the morning.

We have them now, just not passenger carrying. Military drones are autonomous.

They are remotley piloted. https://www.youtube.com/watch?v=_w5NGKt71k4

Engine failure on takeoff is tricky to deal with. See Sullenberger landing on the Hudson for example https://en.wikipedia.org/wiki/US_Airways_Flight_1549

Isn't that a situation where a machine would excel? One challenge is the abundance of decisions with only one non-deadly option, the other challenge is lack of time. The first challenge is solved by well-defined decision trees and mental checklists, but those make the second challenge harder. Computers don't need to balance between checklist thoroughness and decision latency, they can look at all the data they have in practically zero time.

Mandatory bad car analogy: a driver will often prefer a faster response over spending time to check the rear view mirror before slamming the brakes, a computer would never have reasons to do so, they can check without meaningful delay.

No this isn't a situation where a machine would excel because it can't handle unexpected failure modes for which is hasn't been programmed. Human pilots can often come up with a "good enough" solution based on understanding fundamental principles without having any prior experience. For example, the pilots on US Airways flight 1549 intentionally changed and disregarded parts of the checklist for handling a dual engine failure because it wasn't a perfect fit for the actual conditions at the time.

Yes I agree.. As long as the system can simulate the degraded plane, and is fast enough, it can brute force through a huge number of decision trees and scenarios in milliseconds to find the optimal. A human would do something similar but will prune the tree much more heavily than the computer as the human is slow. This is analogous to any chess engine, the difference is that the plane flight requires a much more difficult simulation scenario compared to a chess board. The system has to understand or at least measure what has degraded as well which might be difficult.

(I don't mean to make it sound as I think it's easy to solve, but the basic strategy to solving it is simple :)

The basic strategy is simple, but implementing it is not. I think a perfect real world example is Tesla's recent roll out of their summon feature. The basic strategy of having your autonomous car pull out of a parking spot and pick you up at the door is simple, but the actual implementation shows how difficult that actually is.

There's probably a wide range of unusual, one-off edge cases that almost never happen which would be impossible or difficult to find.

And maybe catching/handling them properly would involve more sensors, when people really just want a well-trained human who is able to improvise in the cockpit.

It would be difficult code a machine to find a good landing spot in extremely variable terrain. For instance, I saw a video of a plane safely crash-landing landing on a four-lane highway with traffic. It would be impractical to code something to handle that very rare situation.

I'm not a pilot, but the by far most dangerous time to have an engine problem is the period that starts the moment you have accelerated too much to safely stop the plane on the ground and that ends with having sufficient speed and altitude to glide back to an emergency landing. There surely are defined procedures for reaching the least deadly outcome for every phase within that period (basically a lookup table based on accumulated energy, with different input options, with either time or airspeed and altitude is proxies for accumulated energy). Being well trained just means applying those rules with minimum hesitation.

Even experience cannot be much of a factor on top of that, as very few pilots will ever experience a scenario like that outside the simulator. My layman's suspicion, honed in the MCAS debates, tells me that the only other thing experience and human "general intelligence and gut feeling" can add to the equation is educated guesswork regarding which data to distrust when things just don't add up consistently due to a malfunction.

Yup:

> V_1 is the critical engine failure recognition speed or takeoff decision speed. It is the speed above which the takeoff will continue even if an engine fails or another problem occurs, such as a blown tire.[9]

* https://en.wikipedia.org/wiki/V_speeds

The plane has to keep going until V_2 (or V_2_min).

>very few pilots will ever experience a scenario like that outside the simulator

One thing that no doubt helped with Sullenberger was he was a hobby glider pilot where that kind of situation comes up regularly if in a much less dangerous way as gliders are slow and light.

This is what companies like Lilium are aiming for [1], and they're aiming for passenger operations starting in 2025 on a "taxi" type basis, probably initially with a pilot on board, though from what I understand this is largely because they don't expect to get approval for autonomous operation from day one. Of course there's no guarantee they'll meet that target, but they're also just one of many companies working on this.

[1] https://lilium.com/newsroom-detail/taking-to-the-skies-fligh...

There's been discussions for years on whether or not the level of automation in modern aircraft causes pilot training to lapse due to misuse. Self-driving planes are much easier than self-driving cars because there is empty space on all sides of the plane pretty much at all times. Autopilot is already generally speaking "a thing", with the exception of takeoff and landing.

Bear in mind even in takeoff and landing scenarios, unlike a public roadway, the airplane has pretty exclusive use of the runway as well.

Pilots are very unlikely the primary cost of aircraft trips. Things like jet fuel are probably much more of the cost driver.

>I feel like we could easily get there before we got self-driving cars, if we wanted to.

The environment has shown quite clearly that we don't want to. We'd probably already have it if we did.

The safety design of aviation relies on having multiple highly competent people in the loop on every flight and exercising their judgement on dozens of things. And that's for the best-case scenario, they also have to be capable of choosing and executing fallback plans when part of the overall system (the plane, air traffic, radio navigation, flying conditions, ground conditions, etc...) degrades or enters an unexpected state.

The technology already exists, it's a much easier problem than self driving cars. Spacecraft can fly themselves to other planets and return. That is a massively more complex problem.

It's a problem of trust, and regulation, not a technical one.