Planes are already fully automatic. Not just in cruise control, but some planes have extra functionality you can pay for to automate takeoff and landing. I’m not sure if they are safer than a human pilot landing in all conditions but I imagine it won’t be long until they are basically equivocal in abilities. Whether or not the role of the pilot continues to exist probably depends more on unions and regulations than if the technology is sufficient, like many jobs in a similar obsolete-yet-perpetuated state today.
But when sensors fail they relinquish control to the pilot. Adding additional sensors could reduce the need for human backup - I wonder if this company added extra redundancy.
I saw a video from one of those "air traffic review" type channels just the other day where a plane was low on fuel and had a malfunctioning ILS and no way to reach somewhere where they could land visually without fudging the rules a little. They landed, but it was dicey.
Diamond Aircraft autoland is meant to work in emergencies like when the pilot becomes incapacitated and the passengers need to initiate it to get to an airport safely.
Correct, but that’s with a perfectly functioning aircraft.
The problems come when you have engines out, landing gear issues, or malfunctioning equipment.
Look at the recent Waymo accidents. Two Waymo vehicles ran into a vehicle being towed backwards. An edge case that a human could have easily avoided, but software couldn’t.
> Look at the recent Waymo accidents. Two Waymo vehicles ran into a vehicle being towed backwards. An edge case that a human could have easily avoided, but software couldn’t.
The complexity of flying an airplane autonomously is way lower than for a car.
The biggest reduction comes from there being (almost) no morons in airspace. Planes are expensive, pilot school is expensive, so the chance of encountering someone doing something they're not supposed to do is way lower - in a car, you have to deal with anything from random little children jumping onto the streets over drunk/tired/phone-using drivers to construction sites.
Then there is the requirement of planes to carry ADS-B transmitters in many airspace classes, so planes precisely know where they are and where everyone else is. And distance between planes is also way larger, which makes avoidance easier.
Basically, as long as the aircraft itself is working fine, everything from takeoff to landing can be done autonomously - and appropriate systems have been developed and used for decades, the first CAT 3 autoland capable aircraft arrived in the late 60s [1]. The only two unsolved challenges are taxiing and in-flight equipment failure... the former can be solved by having a tug pick up a plane on the runway, it's only the failure cases that are a real headache because, at least for now, that requires having highly trained pilots on board.
Autoland is regularly used in CATIIIa/b ILS approaches when weather does not allow for the pilots to transition between instruments and visual flight (otherwise a requirement for the last phase of the approach and the actual touchdown).
Except talking to ATC. Communications are formally structured in 98% of cases, but in the rest 2% they are very free-form / subjective / implying. If you fly a lot, those 2% happen all the time.
Couldn't we create specialized airports that are only for autonomous cargo planes. Then the UAVs can just speak to each other and the automated ATC in a formal protocol? Not a pilot here so I might be missing the nuances.
You're still going to have shared airspace away from the airports. Control is looser away from 'congested areas', of course. But UAV cargo that can only fly outside of congested areas is maybe not very useful.
Sure. The first N years you'd want to have an operator (pilot) onboard to handle edge cases. But... if you already have an operator, you can fly to any existing airport, and analyze the unplanned autopilot disengagements. Thus, you can start right now, no need to build new airports.
When I first flew as a kid I could tell if the pilot trained in the Air Force or Navy. Navy pilots always touched down right on the numbers and often stopped so fast they needed to apply throttle to get to the first turn off - which is the right way to fly if you are landing on a carrier. By contrast air force pilots touched down farther down the runway and coasted to the last turn off - which uses less fuel when you have a long desert runway. Takeoffs Navy pilots would spool the engines up while turning onto the runway and releasing the brakes, while air force pilots would get on the runway and start moving before applying throttle (for take offs there wasn't nearly as much difference as the above makes it seem like).
These days you cannot tell though. All pilots fly the same - I'm sure they use less fuel which is a big deal to the airlines. I'm guessing they are all using the computer to do the flying. which isn't to say they are using auto land, just there are other things in control ensuring they don't do the fuel wasting things anymore.
Once someone points it out it wasn't hard to figure out. There is a big difference in how the Navy and Air force train pilots, relating to their very different missions. The Navy needs pilots that can land on a carrier at sea, and so their training even when on land is designed to get you ready for that. The air force doesn't have to do that, and so they have no reason to teach those skills. Both programs produce highly skilled pilots that can safely fly airplanes, but they have noticeable differences in details.
The US Air force is the worlds largest air force. The US Navy is second largest. While it is possible to learn to fly in ways other than joining one of the above two (Marines, Army, Coast guard, or private schools), they together train much less pilots and so odds are I never had one. I wouldn't have been able to tell these others from an air force pilot most likely.
I can't remember what my pilot friends told me but I'm pretty sure different US airlines prefer to hire pilots trained by different branches. I think Delta has mostly Navy trained pilots and United is more Army/Airforce? Or the other way around, I don't remember the details.
I could believe that. I knew way more about planes, aircraft carriers, etc when I was a kid than I do now (I forgot most of it, and I'm also virtually certain I didn't know this specific fact anyway).
I also used to have like a detailed chronological knowledge of the general troop movements and major battles of the civil war, and now I can barely remember that Gettysburg was after both Bull Runs, and that the first battle was just some Union dudes holed up in a fort getting shelled for days somewhere in the south off the coast. I can't remember what state even, lol.
I definitely remember the first time I noticed a truly different landing in a commercial plane. I can only describe it as "coming in hot": we approached at a higher speed, descended faster, and the reverse thrusters were very active. We stopped really quickly. It definitely felt like the pilot was applying some specialized training, which was confirmed by the pilot who was standing at the doorway to the flight deck when we passed by.
I would bet that the culture started back in WWII when the navy started to use converted passenger ships on Lake Michigan to do carrier certification. They were only about 2/3 as long as a real carrier, so you better follow the landing signal officer's instructions perfectly.
USN carrier landing procedure is to get nose up to a precalculated angle, get speed to precalculated value, then match velocity vector HUD icon to the touchdown point using throttle. Technically, this USN method is not just a special carrier landing procedure, but the correct procedure for all aircraft landings. Airbus passenger planes are flown in basically the same way. I think most piloting textbooks has it within first 1/3rd of Volume 1 or somewhere around.
Thought it cool when I found it, fired up a random flightsim, it works. It's just better. Even takes off mental loads. I can't speak to actual, human carrying flights, but for sim hobbies I'd "recommend" sticking to this method.
e: forgot to add back disclaimer - am not a pilot any shape or form
I'm sure the technology will work fine when everything is perfect. But much like autonomous cars, when things are imperfect they will probably start to fall apart.
The automation relies on ground equipment that isn't guaranteed to be operational and airports aren't required to consider it critical equipment. They can still operate as an airport even if some of these systems are disabled.
The plane isn't magic, you need to beam specific radio signals at it, from the ground, so it knows what to do.
Accommodations for human pilots account for a small percentage of the mass and volume of a cargo plane, but I wonder how differently cargo planes might be designed if they went full drone.
They would probably centrally locate the autopilot and eliminate a lot of wiring for sensors and controls that have to be routed forward to where a pilot normally sits. You could then put a really big cargo door on the front and on the back and load while you unload.
Are military drones not flying autonomously for at least part of the time already? I wonder how much incremental value there is for end to end autonomous vs remote operated flights where a human is involved just for tricky parts.
Are these actually autonomous (no pilot at all) or just very large drones?
https://www.xwing.com"Our system changes the role and location of the pilot. The aircraft is fully self-sufficient with the pilot remotely supervising from a ground control station."
So, there's a pilot somewhere, but it's not clear if the pilot is flying remotely or just there for "backup".
Particularly considering that pilot salaries are going to massively plummet when one superviser can oversee dozens of planes, simultaneously [perhaps].
I think the distinction between remote piloting vs. fully autonomous is a continuum rather than discrete. It's clearly remote piloting if the remote operator has a joystick and all other flight controls that are controlling the aircraft directly in real time. It's clearly fully autonomous if the remote operator issues an instruction that says "take-off from SFO and land at SEA". But there are intermediate cases like the remote operator only issuing way points, or the remote operator having access to some flight controls but not others and only using them in certain situations.
I want to say that “remoteness” versus “autonomy” are actually two separate axes, since even local pilots can in principle defer many flight instructions to an autonomous system. That said, I’m not sure the “remote but non-autonomous” quadrant actually makes sense here.
I think the distinction between remote piloting vs. fully autonomous is a continuum rather than discrete.
Absolutely, I just didn't see much detail on where on the continuum this project sits.
I did find a video that appears to have the pilot just monitoring, but not flying, the plane. A little but worrying, given a similar size/type of plane just crashed by my house (commuter airline at Dulles Airport). If there was no pilot actively flying it (remote or not), would it have been able to safely ditch on the highway? Or would it have crashed into a row of homes instead?
I think it's also worth remembering that commercial airlines already fly on autopilot most of the duration. Pilots fly takeoff, landing, and unexpected scenarios, the rest of the time they're just monitoring the system.
Here's some fun tidbits:
Autopilot landings are actually preferred in bad weather (where airport is equipped).
> On a typical commercial flight, autopilot is used for almost 90% of the flight.
> The autopilot can be engaged as early as 5 seconds after lift-off
> autopilot must be used when flying above 29,000 feet
The article said 2800 miles of testing? Was that a misprint - a single one-way trip from San Francisco to NY is about that distance? Does the FCC or the Air Force get final say in this maybe a little more testing? I AM aware there are drones so maybe a lot of that technology transfers over BUT then there should be a lot more than 2800 miles of flight time data applied no?
> a single one-way trip from San Francisco to NY is about that distance?
You can't go from SF to NY while staying in California airspace afaik. That being said the distance doesn't matter much, I'd be more interested in take-offs/landings
The Air Force gets the final say on their own operations because they aren't legally subject to FAA rules (the FCC only comes into the picture for spectrum use by data links). But they operate under a Memorandum of Understanding and closely consult with each other to minimize risks. Military uncrewed aircraft are typically restricted to specially designated airspace because they can't safely mix with regular civilian traffic (no ability to talk directly with controllers, no "see and avoid" for VHR traffic).
I think the NTIA owns some bands, the FCC others. If the USAF wants to use an FCC regulated band, they talk to the FCC. For instance, a base radio/tv station that broadcasts on a standard frequency.
Distance is anyways irrelevant to the conversation. A commercial airplane can fly itself on a fixed path while at cruising altitude basically forever. It's the takeoff, landing and unexpected situations that you have to worry about.
And clearly with cargo planes, the acceptance of the risk of a crash from an unexpected situation (over an unpopulated area) is a lot higher than if people were on board.
Heh, perhaps we'll end up with inverted-ETOPS, where autonomous airplanes need certifications to spend more than 15, 30, 60 minutes over populated areas (or not water).
Autonomous aircraft require a more stringent MEL - the list of aircraft systems that must be working before you take off, and a more stringent reading of NOTAMS - the list of problems in the air and at the airport that the pilots need to be aware of.
A more stringent MEL means tougher maintenance requirements, which means more mechanic hours per flight hour (in aggregate, on average).
As far as I know, it's difficult to get a computer to read all the NOTAMS in an automated way that will consistently produce a useable plan with no missed problems. It would need to have no false positives and no false negatives. What am I allowed, What am I forbidden.
Of course people have the same problem, but they have problem solving skills and can decide what weight to give any issue that crops up.
There are also a lot of assumptions about what a pilot can work around that will have to be understood.
> As far as I know, it's difficult to get a computer to read all the NOTAMS in an automated way that will consistently produce a useable plan with no missed problems.
To be fair, it's extremely difficult for humans too, and incidents have happened due to this in the past:
> A more stringent MEL means tougher maintenance requirements, which means more mechanic hours per flight hour (in aggregate, on average).
Probably true, but I wonder if it just becomes easier to hire/train mechanics than it is to do the same with pilots, especially as flight hour requirements ramp up.
The sweet spot for autonomy might be very long fights, where you're replacing multiple pilots. Even replacing all but one on a long haul flight might be a huge boon
Cessna Caravans are basically free as far as military aircraft, and for missions like fetching food or fuel, you can take some risks. I am no pilot but the stringent MELs and deciphering NOTAM risk can be relaxed or better controlled as these autonomous cargo systems get real world reps in. Also, simply holding up 100 or so fully loaded caravans, waiting for a break in the weather to dispatch, seems easy to envision! 5 remote pilots overseeing a few takeoffs at a time vs 100 pilots idling to begin their round trip journey
and for missions like fetching food or fuel, you can take some risks
When flying through congested California airspace?
A Cessna Caravan just crashed at my local airpot. If there wasn't a pilot on board flying, would it have been able to successfully ditch on an empty stretch of road with no injuries (and barely any damage to the plane)? Would a remote pilot have been able to do the same? An AI pilot? I have my doubts about the latter.
I mean I get the fully autonomous flight is interesting and should be pursued, buts laughable that this is billed as a cost savings. Of course flying a $2-3 million aircraft is going to be cheaper than flying a $100 million one, pilot cost is negligible in that equation
The two aren't independent, though. When you have to pay hundreds of thousands of dollars for humans to operate an airplane, you might as well get the most out of the fixed cost of the airplane to begin with. If you don't have to do that, you can build cheaper smaller airplanes to fly cargo on routes that wouldn't be economical at human operation scale.
You can mass produce an autonomous fleet without consideration for a training pipeline. (You can also lose a lot of life-support systems, and engineer the plane to be more disposable in combat.)
Definitely true - but what is that in the context of a ~$100m c-130? 1-3%, not nothing, but you probably gain more with engine upgrade programs, adding blended winglets etc.
> what is that in the context of a ~$100m c-130? 1-3%
For the C-130, I’d guess 10%+ on operating costs based mostly on increased utilisation.
For a new vehicle, support and redundancy alone should yield 10%+. Doubling or tripling production, based simply on better amortisation of development costs, zero learning curve assumed, another 20%+. Then add on reduced operating costs due to lack of pilot training and pay as well as higher utilisation.
Again - this makes sense when you are talking about a ~$50/k a year driver driving a $50k car, a lot less when you are talking about $400k of pilot salaries when you are spending ~$12m a year operating a ~$100m c-130
Was this supposed to be one of many tests, the final test, or THE test? I’ll read it again, but
the way the writer has it framed makes it appear as if the Air Force conducted a 2800 mile flight-test and said, “Alright, we’ve seen enough,” and called it good.
[quote]
“The use of Xwing’s autonomous aircraft eliminated the need to
fly a larger aircraft such as a C-130 to deliver critical cargo to the warfighter on short notice,” said Maxime Gariel, president, chief technical officer and co-founder of Xwing.
[/quote]
Did it though? All jokes aside, since when was California an active war zone? (I tried to frame this better—-under siege, in a fire-fight, etc. but kept coming up short)
This would be too far a stretch to assume the average skimmer to connect, but as a side-thought, perhaps we should look at it as this marking the beginning of the experiment, rather than the end.
I think the idea is cool and if it can be extended to private applications such as medical and critical, all while lowering costs and increasing access to the general public, that’d be great.
But reading this left me with more questions than answers
Yes, I am left believing this is the first test and not the announcement of a contract for ongoing services.
It's potentially very economical, in that there is a shortage of pilots. Reducing flight hours required for low tonnage short haul flights is potentially going to save your experienced pilots for more critical missions
> All jokes aside, since when was California an active war zone?
Warfighters aren't always based in active war zones. For example, the last round of U.S. strikes against targets in the Middle East involved non-stop bomber flights from the US.
I've never thought – though it's obvious now – that if a drone is capable of flying remotely in the middle east, it's just as capable of flying remotely in the US.
Legitimate qualms about drone strikes aside, it seems like a big step forward to not have to send a crew with a plane for cargo transportation. Hopefully this can be the case for private cargo transport, in time.
92 comments
[ 3.8 ms ] story [ 210 ms ] threadBut when sensors fail they relinquish control to the pilot. Adding additional sensors could reduce the need for human backup - I wonder if this company added extra redundancy.
edit: found it! https://www.youtube.com/watch?v=HkzVMZFWduc
The problems come when you have engines out, landing gear issues, or malfunctioning equipment.
Look at the recent Waymo accidents. Two Waymo vehicles ran into a vehicle being towed backwards. An edge case that a human could have easily avoided, but software couldn’t.
The complexity of flying an airplane autonomously is way lower than for a car.
The biggest reduction comes from there being (almost) no morons in airspace. Planes are expensive, pilot school is expensive, so the chance of encountering someone doing something they're not supposed to do is way lower - in a car, you have to deal with anything from random little children jumping onto the streets over drunk/tired/phone-using drivers to construction sites.
Then there is the requirement of planes to carry ADS-B transmitters in many airspace classes, so planes precisely know where they are and where everyone else is. And distance between planes is also way larger, which makes avoidance easier.
Basically, as long as the aircraft itself is working fine, everything from takeoff to landing can be done autonomously - and appropriate systems have been developed and used for decades, the first CAT 3 autoland capable aircraft arrived in the late 60s [1]. The only two unsolved challenges are taxiing and in-flight equipment failure... the former can be solved by having a tug pick up a plane on the runway, it's only the failure cases that are a real headache because, at least for now, that requires having highly trained pilots on board.
[1] https://en.wikipedia.org/wiki/Autoland
But drones fly, and this is a drone.
"SkyNet: We do logistics!"
These days you cannot tell though. All pilots fly the same - I'm sure they use less fuel which is a big deal to the airlines. I'm guessing they are all using the computer to do the flying. which isn't to say they are using auto land, just there are other things in control ensuring they don't do the fuel wasting things anymore.
The US Air force is the worlds largest air force. The US Navy is second largest. While it is possible to learn to fly in ways other than joining one of the above two (Marines, Army, Coast guard, or private schools), they together train much less pilots and so odds are I never had one. I wouldn't have been able to tell these others from an air force pilot most likely.
I also used to have like a detailed chronological knowledge of the general troop movements and major battles of the civil war, and now I can barely remember that Gettysburg was after both Bull Runs, and that the first battle was just some Union dudes holed up in a fort getting shelled for days somewhere in the south off the coast. I can't remember what state even, lol.
Reading books
https://www.youtube.com/shorts/KLAOPg6AxFM
[1] https://www.youtube.com/watch?v=eDR8znA1kxw
Thought it cool when I found it, fired up a random flightsim, it works. It's just better. Even takes off mental loads. I can't speak to actual, human carrying flights, but for sim hobbies I'd "recommend" sticking to this method.
e: forgot to add back disclaimer - am not a pilot any shape or form
https://www.usgs.gov/media/images/geophysical-survey-airplan...
They're not all as high and angled as the image above.
The plane isn't magic, you need to beam specific radio signals at it, from the ground, so it knows what to do.
Take out the humans and add remote piloting and they'll be flying bombs just waiting for hackers to take control of them.
Though I also wonder how it takes ATC direction.
Better to use drones, which even if it fails 10-20% of the time is better than not failing to supply troops on the ground at all.
https://www.xwing.com "Our system changes the role and location of the pilot. The aircraft is fully self-sufficient with the pilot remotely supervising from a ground control station."
So, there's a pilot somewhere, but it's not clear if the pilot is flying remotely or just there for "backup".
Perhaps a better idea would be to have two pilots, together, monitoring dozens.
Absolutely, I just didn't see much detail on where on the continuum this project sits.
I did find a video that appears to have the pilot just monitoring, but not flying, the plane. A little but worrying, given a similar size/type of plane just crashed by my house (commuter airline at Dulles Airport). If there was no pilot actively flying it (remote or not), would it have been able to safely ditch on the highway? Or would it have crashed into a row of homes instead?
Here's some fun tidbits:
Autopilot landings are actually preferred in bad weather (where airport is equipped).
> On a typical commercial flight, autopilot is used for almost 90% of the flight.
> The autopilot can be engaged as early as 5 seconds after lift-off
> autopilot must be used when flying above 29,000 feet
https://travelradar.aero/the-importance-autopilot-commercial...
You can't go from SF to NY while staying in California airspace afaik. That being said the distance doesn't matter much, I'd be more interested in take-offs/landings
But I could be wrong.
The 2800 miles was multiple flights within California between Jan. 22 and Feb. 4.
This article has more detail:
https://www.flyingmag.com/xwing-awarded-military-airworthine...
A more stringent MEL means tougher maintenance requirements, which means more mechanic hours per flight hour (in aggregate, on average).
As far as I know, it's difficult to get a computer to read all the NOTAMS in an automated way that will consistently produce a useable plan with no missed problems. It would need to have no false positives and no false negatives. What am I allowed, What am I forbidden.
Of course people have the same problem, but they have problem solving skills and can decide what weight to give any issue that crops up.
There are also a lot of assumptions about what a pilot can work around that will have to be understood.
To be fair, it's extremely difficult for humans too, and incidents have happened due to this in the past:
https://fixingnotams.org/wp-content/uploads/2019/11/Field-Gu...
Probably true, but I wonder if it just becomes easier to hire/train mechanics than it is to do the same with pilots, especially as flight hour requirements ramp up.
One human can review reports remotely for dozens of planes compared to an in-the-cockpit pilot.
When flying through congested California airspace?
A Cessna Caravan just crashed at my local airpot. If there wasn't a pilot on board flying, would it have been able to successfully ditch on an empty stretch of road with no injuries (and barely any damage to the plane)? Would a remote pilot have been able to do the same? An AI pilot? I have my doubts about the latter.
Automated systems can ‘fail’ systemically and completely across the board, which changes the risk profile.
I.e. Pilots won’t all crash loop at the same time if the wrong kind of character happens to be in a NOTAM. Probably.
You can mass produce an autonomous fleet without consideration for a training pipeline. (You can also lose a lot of life-support systems, and engineer the plane to be more disposable in combat.)
For the C-130, I’d guess 10%+ on operating costs based mostly on increased utilisation.
For a new vehicle, support and redundancy alone should yield 10%+. Doubling or tripling production, based simply on better amortisation of development costs, zero learning curve assumed, another 20%+. Then add on reduced operating costs due to lack of pilot training and pay as well as higher utilisation.
[quote] “The use of Xwing’s autonomous aircraft eliminated the need to fly a larger aircraft such as a C-130 to deliver critical cargo to the warfighter on short notice,” said Maxime Gariel, president, chief technical officer and co-founder of Xwing. [/quote]
Did it though? All jokes aside, since when was California an active war zone? (I tried to frame this better—-under siege, in a fire-fight, etc. but kept coming up short)
This would be too far a stretch to assume the average skimmer to connect, but as a side-thought, perhaps we should look at it as this marking the beginning of the experiment, rather than the end.
I think the idea is cool and if it can be extended to private applications such as medical and critical, all while lowering costs and increasing access to the general public, that’d be great.
But reading this left me with more questions than answers
It's potentially very economical, in that there is a shortage of pilots. Reducing flight hours required for low tonnage short haul flights is potentially going to save your experienced pilots for more critical missions
Yes, that's what they were doing during the test.
> All jokes aside, since when was California an active war zone?
Warfighters aren't always based in active war zones. For example, the last round of U.S. strikes against targets in the Middle East involved non-stop bomber flights from the US.
Legitimate qualms about drone strikes aside, it seems like a big step forward to not have to send a crew with a plane for cargo transportation. Hopefully this can be the case for private cargo transport, in time.