A plane makes an autonomous landing | Hacker News Reader

A plane makes an autonomous landing

techcrunch.com

35 pointsprostoalex32 comments

The article sort of disses traditional landing guidance systems but misses the point of these. Specifically that such instrument approaches are used because one can’t see the runway at all. Thus a system that relies on cameras to land is academically interesting but not all that practical. Also with the advent of GPS-WAAS based LPV approaches one doesn’t need equipment on the field and a high precision 3D approach path can be created anywhere.

As for replacing pilots for “autonomous” aircraft a pilot friend of mine said to ask actual pilots if they’d get on board a fully self flying aircraft and you’d be hard pressed to find one. As he put it, the automation on planes is amazing and a big help but it does mess up and when it does the automation can crap out big time. For the foreseeable future there’s still no replacement for a qualified pilot at the controls ready to take over.

Finally there’s a lot of debate in the aviation community that too much automation actually decreases safety because pilots not flying enough manually lose skills. See the SFO accident with the Asiana airliner that crashed for no other reason than the ILS was down so they couldn’t fly with the autopilot. They crashed short of the runway on a simply manual landing in perfect weather because they couldn’t fly the plane properly!

The landing in the article was a really cool experiment though.

I'd fly hundreds of domestic flights on an un-fixed 737 Max before I'd fly a single flight on a pilot-less airplane at this point in the state of technology.

s/hundreds/thousands/

same for self driving cars. software quality is too low.

They weren't suppose to take "move fast and break things" literally!

(It's really weird how "literally" has two literally opposite meanings.)

Entertaining 1 minute on "literally" https://www.youtube.com/watch?v=4-ImRMJX68s

> As for replacing pilots for “autonomous” aircraft

My fear, is that AI would generally work, but do worse than humans in failure scenarios. Flying is inherently higher-risk than cars because cars can (almost) always just stop moving and be safe.

> They crashed short of the runway on a simply manual landing in perfect weather because they couldn’t fly the plane properly!

This. It was a clear day, but they couldn't read the visual approach indicators and didn't have the sense to go around.

> As for replacing pilots for “autonomous” aircraft a pilot friend of mine said to ask actual pilots if they’d get on board a fully self flying aircraft and you’d be hard pressed to find one.

That doesn't say much - pilots are only experts in existing autoplilot systems designed to have human backup, not future autopilot systems designed to not have human backup.

> pilots are only experts in...

Since you are questioning the judgement of pilots based on their expertise, I am going to present my qualifications:

I am a private pilot with an instrument rating and over 900 hours and 23 years of experience. I was also an AI researcher in the first half of my career. I wrote my Ph.D. thesis on how to build autonomous vehicles that would navigate through natural terrain (I did this while working for NASA on the research program that led up to the first Mars rover mission). I was also the PI for the Remote Agent Executive [1], which controlled the first (and to my knowledge still only) fully autonomous interplanetary spacecraft. So I think I'm about as qualified as a person could be to make the following judgement: I would absolutely not fly in a fully autonomous spacecraft. No fucking way. I have no doubt that they would work fantastically well >99% of the time as long as everything was going to plan. But I don't want to be in the aircraft the one time that something goes wrong that the system is not equipped to handle, and watch my life slip away watching the aircraft doing something stupid because some minimum wage sub-contractor in Asia forgot to put a semicolon in the right place.

And I don't think you need to be an expert to realize that full autonomy is a Really Bad Idea. All you have to do it look at how the 737MAX MCAS debacle is unfolding. MCAS was autonomy writ very, very small, and 346 people are dead.

[1] https://ti.arc.nasa.gov/tech/rse/vandv/remote/

> I would absolutely not fly in a fully autonomous spacecraft. No fucking way.

So I guess you'd never get into a self driving car- the problem of driving a car on a trafficked road seems enormously more complex than that of piloting an aircraft.

I would have serious reservations about using a self-driving car. And a self-driving car without a steering wheel that would let me take over? No. Absolutely not.

I drove a rental car with lane-keeping for the first time the other day, and even that had some failure modes that made me queasy.

>And I don't think you need to be an expert to realize that full autonomy is a Really Bad Idea. All you have to do it look at how the 737MAX MCAS debacle is unfolding.

Thank you for this. Boeing's hard on for de-skilling pilots isn't getting nearly enough credit as the reason why all those people died.

Totally incorrect title by techcrunch. This is in no way the first, not by three decades.

Russian SPACEplane makes autonomous landing in massive crosswind: (1988) https://en.m.wikipedia.org/wiki/Buran_(spacecraft)

Huh? Not really, Buran used instrument approach and landing, assisted by the Vympel-N system on the ground, just like Shuttle or any other aircraft. The only relatively innovative thing in the Buran EDL sequence was the fully automated decision making procedure which enabled the divert maneuver. (you're probably thinking of this when calling it autonomous)

It's not clear from your reference (and I've read a few of the further references) how Buran determined where to land exactly. Was there really no signalling from the ground present, as in the featured article?

There are a lot of people in this thread putting down the idea of an autonomous plane, but frankly, as a pilot it is a much easier environment to automate than driving.

In the world of instrument flying there is very little to hit which isn't clearly charted, you have an external navigator to keep you clear of most other planes (Air Traffic Control), and the landing environment is kept very normalized by the hard work of the FAA. This is orders of magnitude simpler to pull off than self driving cars, and it might save the General Aviation industry.

The problem is handling failures. We could do automatic landings back when the British introduced Autoland, and in fact any serious autonomous landing will use ILS as the backbone. The problem is that we barely got to the point where an autonomous plane was capable of reasoning enough to do a go-around, but I have zero belief that it would handle failures in other ways than "crash in unoccupied terrain" (if there's any).

Here you can find the initial press releases and the demonstration videos

press releases: https://magazin.tu-braunschweig.de/pi-post/sichere-landungen... https://www.tum.de/nc/en/about-tum/news/press-releases/detai...

Demonstration videos: https://www.youtube.com/watch?v=tPFC7Ki2JT0 https://www.youtube.com/watch?v=KAnzcNrenCU

"Now it would be natural to think that with the sophisticated autopilot systems that we have today, a plane could land itself quite easily. And that’s kind of true "

Indeed, I believe a substantial portion of commercial flights are actually landed automatically (contrary to what the article claims).

Bravo for doing it autonomously though, it's obviously the next step.

Contrary to popular belief, autoland systems are not used that frequently, at least in the US. Typically the only time autoland is accomplished is when the pilots or aircraft need a practice run for currency or weather conditions actually dictate its use. Autoland is required for CAT II and III approaches on the aircraft I operate. CAT II weather minimums are 100-200 ft decision height and 1,800 RVR for many operators. CAT III is even lower. If the weather is greater than that autoland typically is not going to be used.

This system is very interesting though for small aircraft. As the article states, small aircraft often fly into small fields that may not have as much precision landing equipment on the runway. I could see this system being an excellent augmentation system or an additional input in the pilot’s planning of the approach and landing.

[EDIT] clarity about CAT II and III

> 50 ft decision height

What does that mean?

The height at which the pilot has to abort the landing attempt if they can't visually confirm that they're correctly set up for landing.

I was actually mistaken. That’s for a CATIII. CAT II is actually 100-200ft. That’s how low the reported cloud bases can be.

Small aircraft also tend to have less experienced pilots, so more help managing the complexity of the landing could reduce incidents like this one in Houston (stall leading to crash during go around):

https://www.flyingmag.com/aftermath-cirrus-crash-june-2016/

You can tell/guess sometimes. An auto landing would feel kind of rough. The reason for this is that a rough landing removes enough energy to make a bounce unlikely and that would be undesirable as recovering from those is a bit tricky. The gear can handle it. So they intentionally plant it down firmly instead of "greasing it in".

Takeoffs and landings are pretty much the only point where commercial aviation involves any manual flying at all these days.

Autonomous flight will likely go through years of testing and scrutiny before it will be considered safe. But technically it's not fundamentally any harder than a self driving car. I could see this help reduce work loads for pilots short term.

There is a huge fundamental difference - you can't stop on the curb and send SOS when something breaks.

There's no margin for error.

Of course there are margins for error. They're just smaller.

The article doesn't spell out why Category IIIc (the mode in which the "decision height" is zero and the computers will put the plane down even if there is literally no visibility at all) isn't used in practice. The reason is: ok, now your plane is on a runway, we all agree that's good, but what next? You can't see, that's why you need Category IIIc rather than just Category IIIb which is sometimes done in bad weather. So you can't taxi the plane off the runway safely. Your ground controllers can't see you either. This is strictly better than not being able to land, but that's all.

So Cat IIIc (and to some extent even other Cat III landings) come into their own when the busiest airports can safely handle ground operations in all weathers, which may never happen.

I’m not sure how useful this is when the weather visibility is poor, there’s an obstruction on the runway or it’s dark. But what could be useful is using this to serve as an additional alert for when a pilot is performing a bad landing. For example, could this have been helpful in the Asiana flight 214 crash at SFO?

Thermal imaging systems can typically get better visibility through the moisture. I do agree, however, this certainly won’t be replacing an ILS any time soon.

This system would be an excellent input to the pilot for managing glide path when the ILS system is inoperative. The PAPI (precision approach path indicator) light system also provides guidance but it’s not a very tight path.

How do modern HUDs overlay the runway? For example even the Shuttle displayed a runway outline during landing [1]. Is it just really good position determination or is it computer vision? I assume once you're on the localiser the aircraft has a pretty good idea of the runway geometry.

[1] https://m.youtube.com/watch?v=8W4cfIyNvts

> A plane can land automatically using ILS and other systems, but it’s rare and, even when they do it, it isn’t truly autonomous — it’s more like the airport is flying the plane by wire.

This is a misunderstanding of ILS. ILS provides inputs to the pilot and the instruments on board. It’s not like fly-by-wire at all.

Like usual, I don't even know where to start with the ill-informed HN comments on aviation ...

This was a landing on a blue-sky day in a country, Germany, with almost no airports compared to the USA.

So although handy for a "pinch-hitter" type emergency landing in broad daylight with one airport and one runway in sight, otherwise it's of little practical use.

Also, any landing attempt has 2 outcomes: a landing or a go-around. If the tower says at 50' AGL, "there's a deer on the runway, go around." what would this system do exactly?

Note that some biz jets do have synthetic vision now that can be used in IMC, but not trainers.

Reply on news.ycombinator.com