Surely the robot needs software to run itself *and* land planes? That's double the effort surely?
Robot lands a 737 by hand, on a dare from DARPA
An outfit called Aurora Flight Sciences is trumpeting the fact that one of its robots has successfully landed a simulated Boeing 737. Aviation-savvy readers may well shrug upon learning that news, because robots – or at least auto-landing systems - land planes all the time and have done so for decades. Aurora's excitement is …
COMMENTS
-
-
Wednesday 17th May 2017 09:00 GMT Anonymous Coward
The human world is built around human interfaces. Once you have the design that can see and interpret the visual information that we can, hear the audible clues, and manipulate the world as we do with our hands and feet you have a design that can automate anything a human can control....... probably very very badly at first.
Hopefully from a robotic point of view flying a plane is mechanically no harder than opening a door.
-
Wednesday 17th May 2017 16:01 GMT Anonymous Coward
"Hopefully from a robotic point of view flying a plane is mechanically no harder than opening a door."
Except for the minor point that said robot needs input. They mention machine vision but I'm a bit suspicious about this and unless machine vision has progessed in leaps and bounds in the last few days I'm wondering exactly how many of the dials, screens and switches it can actually read and how much training it required to be able to read tbe ones it can.
Oh , and "sitting in the co-pilots seat" seems to be a euphamism for ripping the seat out and bolting a bot in its place. Hardly a 5 minute job. I imagine it would be somewhat simpler to plug in an upgraded autopilot board that could do the job itself.
-
-
-
-
Wednesday 17th May 2017 07:47 GMT Baldrickk
Re: Huh?
It will still need to be customised for every type of plane, as they are all different with different characteristics - even moreso with a robot rather than fly by wire, as not only do the planes handle differently, the controls will be in different places in different cockpits...
-
Wednesday 17th May 2017 07:52 GMT Anonymous Coward
Re: Huh?
Customised, yes, but independently of fiddling with the fly-by-wire flight control systems themselves. Its not a free job, but far simpler than changing fundamental software components of the aircraft. This is sensible if a retrofit is required.
New aircraft could well have the software built-in and not require "protuberances" to operate the flight controls though.
-
Wednesday 17th May 2017 14:24 GMT JohnG
Re: Huh?
"Customised, yes, but independently of fiddling with the fly-by-wire flight control systems themselves. Its not a free job, but far simpler than changing fundamental software components of the aircraft."
I don't think so. Fly by wire systems already have all the sensor inputs and control outputs necessary to fly the aircraft. All that is needed is some software to use those inputs and outputs, whilst observing flying rules (Try to land on a runway, ideally, the correct one. Don't land inverted. You need wheels to land.) A robot has to use a camera with optical recognition and mechanical manipulators - which will inevitably introduce lag and inaccuracies/errors - and then it has to have the same software to actually use inputs and flying rules to make outputs. Any system would have to be passed for use on each aircraft type.
P.S. In the video, the robot accidentally pushed the control yoke forward, whilst looking at instruments.
-
-
-
-
Wednesday 17th May 2017 10:06 GMT DropBear
Re: Huh?
"I have it on good authority that if you can fly a Sopwith Camel you can fly anything!"
...even a GeeBee Model R...?
-
-
Wednesday 17th May 2017 10:02 GMT DropBear
Re: Huh?
"Well it can't be that hard, meatsacks have been flying different types of aircraft for years."
We can have this discussion again when household robots that can manually wash dishes, operate normal vacuum cleaners and generally fully care for the elderly on their own are ubiquitously common and stupidly reliable. Until then, I'm just going to assume reading this that it's either somehow April the 1st again or lots of important and apparently sane people somehow went start raving mad without anyone noticing.
-
-
-
Wednesday 17th May 2017 14:33 GMT JohnG
Re: Huh?
"It will still need to be customised for every type of plane [...]"
"So? You need to "customise" pilots too. I.e. train and re-train them for every type they are going to fly in."
Yes. You don't have to start from scratch but there are some steps involved in switching between different aircraft types. You can't pass your PPL in a Cessna and fly a 747 the next day.
-
Wednesday 17th May 2017 16:09 GMT SImon Hobson
Re: Huh?
Pedant alert ...
You can't pass your PPL in a Cessna and fly a 747 the next day.
Actually, yes you can if you have the money. I believe it is technically possible to train for and get your PPL in a 747 - though the difference in cost between the per-hour cost of a light piston single and a 747 would make it a very expensive proposition.
Assuming you took the conventional route to your PPL (SEP(A)), you could still jump in a 747 the next day if you had the money to buy lessons for the type-specific qualification.
-
-
-
Wednesday 17th May 2017 17:30 GMT Cynic_999
Re: Huh?
"
It will still need to be customised for every type of plane
"
It could have the same central "autopilot engine" software for all types, but supplied with different fixed parameters for each aircraft type. The physics of fixed-wing subsonic flight are the same for all aircraft from a two-seat Cessna to an Airbus, only the physical constants (and a few variables such as position & quantity of loaded mass) cause the differences in behaviour. Many of the less significant variables could even be learned by the autopilot literally "on the fly" (auto-tuning). No need for visual recognition or cameras either, the autopilot can have its own set of gyro instruments and radio navigation receivers, and will need only external connections to a couple of aerials and a pitot and static air feed (readily available or easily tapped), and maybe the AoA indicator - though that can be computed from other data so is not strictly necessary except as a cross-check.
The big advantage is that you only need one such autopilot to be able to equip any one of a fleet of different aircraft on demand, rather than fitting every aircraft with a full autopilot on the off-chance that it will one day be needed for operational reasons. The mechanical connections can be engineered so as to easily adjust or have adaptors fitted to suit a wide range of cockpit layouts, so installation does not have to take weeks.
-
-
Wednesday 17th May 2017 08:10 GMT Spudley
Re: Huh?
but it is aircraft and manufacturer independent, and can be tested in isolation. much, much easier to deploy, or indeed wrench out of the seat in the event of malfunction. Wrenching is tricky in-flight with software...
The image that springs to mind is Arnie wrenching the JohnnyCab robot.
-
Wednesday 17th May 2017 09:15 GMT Anonymous Coward
Re: Huh?
"no operational advantage I expect."
Worse than that, ignoring the problems of verifying the software running the robot, the actual use of it introduces more points of failure than interfacing directly with the fly by wire system. You now have more possible mechanical points of failure, and the whole thing needs to "fail safe" i.e. it has to be mechanically impossible for it to get stuck irreversibly against the controls in a way which will cause the aircraft to crash. And then of course, there's the *more* complex task of writing the software for the thing.
Being charitable, it's a fun project, and kudos to the team for building it and making it work, but ultimately, I really hope it never makes it into a real aircraft.
-
-
-
Wednesday 17th May 2017 19:03 GMT Mark 85
Re: Huh?
The 737 is a very common plane and was probably easy to "borrow" one for the test. I would think that the fly-by-wire automations were turned off to allow the robot to do it's thing with human oversight. Or perhaps the on-board system was allowed to be active to compare notes, so to speak.
-
Friday 7th July 2017 10:03 GMT Chris 239
Re: Huh?
Acutally watching it the robot used the auto pilot to do the job - the only actual flight controls it used were the flap lever ( because there is no auto flap on this type) and the reverse thrust levers (after it's touched down) every thing else was done with the auto pilot and auto throttle controls (the knobs and buttons just below the glare shield).
-
-
-
Wednesday 17th May 2017 08:17 GMT Brenda McViking
I'm sure we'll see it as a retrofit option. The road to autonomy as we know it involves robotic assistance for the meatbag first, followed by lots of supervised machine learning, and then a gradual phase out of meatbags who are sat there with a hand on the big red override switch. We'll see it with land vehicles first, then marine, then aviation.
Airlines know full well that they have to pay two very expensive pilots salaries when their effective professional utilisation is probably around the 3-5% mark. Most of the time they are staring at instruments of an aircraft that requires no intervention. Even if there is a mandate for human intervention, this could be done in the same way that military drone operations are done now - high reliability datalinks with a professional pilot at the other end, potentially controlling multiple aircraft if the required utilisation rate of the pilot is low. I'm not saying your passenger planes are going to be affected for decades yet, but semi-autonomous cargo? Probably could be done by 2025.
-
Wednesday 17th May 2017 08:40 GMT toughluck
Suppose that, on average, a captain earns $200,000 and the first officer earns $100,000 (these figures are way overstated) and both fly 100 times a year (and that one is way understated).
Each flight would then cost $3,000 in flight deck crew cost.
Fuel capacity of an aircraft varies between ~15 tons for a small passenger jet (737 or A320), and ~70 tons for a widebody like 787.
That's between some 4900 and 23000 gallons of jet fuel. I've searched for current prices of jet fuel and I found wildly different prices, from ~140¢/gal (IATA) to as high as 293-790¢/gal (aviationweek, low-high prices across the US). Going with the lowest figures, it costs from $6,800 to $32,000 to fuel up an aircraft. Flight deck crew costs doesn't even enter into the picture.
-
Wednesday 17th May 2017 11:01 GMT John Smith 19
"Flight deck crew costs doesn't even enter into the picture."
However that represents one recurring cost (pilot salary) versus 1 single payment (with a service contract as well?)
It also (might) represent several kilos of mass IE the 2nd pilot that does not have to be carried and can be replaced by more useful "stuff" (IE something the airline can charge you for rather than a cost they have to pay). TBH that 'bot looks quite heavy but then it's a PoC design and I don't think DARPA said it had to be lighter, just about the same as a pilot.
That said as a permanent installation in a commercial aircraft they could probably simplify the design and have its control boards share the equipment racks with other stuff and take various other measures to cut its weight below that of a meatsack. 10-20Kg below the average weight of a pilot might sound nothing but over the life of aircraft (and with some airlines on <1% profit) that multiplies up to a shed load of cash.
That said autoland systems are certified to 1 fail in 1x 10^9 operating hours. That sounds ridiculous but consider (using round numbers) 6000 737's in operation x 10 mins of autoland operation x # of flights a day --> 1000 hrs x # of flights a day. IOW you've racked up 1 billion operations in 3 years.
Demonstrating this system can have that level of reliability will be tough.
But it's going to happen in commercial aviation at least. The bottom line is (sadly) the bottom line.
-
-
Wednesday 17th May 2017 12:33 GMT John Smith 19
"Not until they invent planes that you only fuel up once and never pay for fuel again."
Sorry, I was not clear.
The single payment I was talking about was the purchase cost of the robot. Beyond X number of flights the saving from not paying for pilot (or co-pilot) pays for the 'bot.
If it's also lighter than the average pilot that can also save a (smallish) number of Kg of fuel per flight, but over the life of the aircraft that can also be substantial. Otherwise they might trade if for something else. It might put new routes within range of existing aircraft for example.
-
-
-
Wednesday 17th May 2017 15:23 GMT Tom 7
RE: Flight deck crew costs doesn't even enter into the picture.
Still wont stop some accountant saving the company money. Saving that $100,000 a year may not make any difference to ticket prices but its a health bonus for someone.
I've noticed in some companies above a certain level saving $100,000 will lead to a much higher bonus. Maths is different in the thin air at the top.
-
Wednesday 17th May 2017 19:13 GMT Mark 85
While I tend to agree, but there's the human thing. Not just from passengers needing reassurance that there's a human in charge. We've had several incidents (for lack of a better term) where the pilot managed to save the plane and passengers a nasty death. On the other hand, we've had a few incidents where not having a human pilot might have been a better thing.
-
-
-
Wednesday 17th May 2017 10:14 GMT DropBear
Re: first they came for the co-pilot...
It actually boggles the mind how we can even talk about attempting to replace drivers with a straight face, while we require our vehicles moving around on fucking rails to still be operated (to the sole extent of "go faster" "go slower" "stop here") by humans, all across the board.
-
Wednesday 17th May 2017 10:22 GMT Nik 2
Re: first they came for the co-pilot...
Partly because train drivers are heavily unionised, and will take strong action to head off this threat, which is why implementation is limited to wholly new systems like DLR in London and to shuttle systems at airports, etc.
Also, I used to know a train driver, and he would have to leave the cab of his train to perform maintenance on a fairly regular basis. AIUI, this was often of the percussive nature.
-
Wednesday 17th May 2017 10:48 GMT Rich 11
Re: first they came for the co-pilot...
And the DLR is comparatively easy to automate, since much of the track is up out of the reach of stupid meatsacks doing silly things after five pints of snakebite and black.
I haven't actually been on the DLR for ten years or so. Does each train still have a guard aboard to stop the stupid meatsacks doing silly things with the doors when they realise they've missed their station?
-
Wednesday 17th May 2017 11:42 GMT Putters
Re: first they came for the co-pilot...
Yes, DLR still has Train Captains. They can drive the trains in emergency / fault conditions. They are first line fault diagnosis and repair. They also provide a degree of security and a visible point of contact for customers.
One of the other requirements for "no crew" trains is that the train is easily accessible in emergency situations - which explains the footway in the tunnel on the DLR when it goes into Bank, and which rules out most of the Tube as being driver free.
Another is Platform Edge doors - which rules out any Lines of the Tube where more than one stock shares a platform as the doors don't line up - so forget the Jubilee, and Met, and District / Met and Piccadilly
-
-
-
-
-
Wednesday 17th May 2017 08:29 GMT Anonymous Coward
Genuinely interesting idea
And kinda though provoking. The advantages of making robots that can use interfaces that humans use, rather than specifically interfacing for a robot is clear. You only need one interface, and the robot could be interoperable for other human interfaces too, perhaps such as a car . However, there is zero, and I mean no excuse whatsoever not to put an airline pilot's cap on the robot in that picture. What the heck were they thinking?
-
Wednesday 17th May 2017 11:12 GMT John Smith 19
"The advantages of making robots that can use interfaces that humans use"
Yes. A group called Shadow Robotics noticed as much about 30 years ago. It turns out that building a human strong, human speed, human weight and at least human accurate robot is very tough.
The big one is the human weight. Human muscle is actually very light for its output relative to other systems and real human hands have 100s of degrees of freedom, enabled by each one.
-
-
Wednesday 17th May 2017 08:51 GMT SkippyBing
All of this has happened before, and all of this will happen again
The UK actually did something similar to this with its Sea Vixen drone programme
https://www.seavixen.org/sea-vixen-drone-d3-era
The difference mainly being they never got round to trying it in different aircraft as far as I can tell and it didn't have a robotic arm on the controls. But the basic principle of a removable package that can replace the pilot in a variety of aircraft is there.
-
Wednesday 17th May 2017 12:49 GMT John Smith 19
Re: All of this has happened before, and all of this will happen again
From the site
https://www.seavixen.org/sea-vixen-drone-d3-era
"So FR came up with something called a Universal Drone Pack (UDP) and all the avionic modules for the remote control were on a rack slid onto the ejector seat rails in the Observers position."
Good point. It had not occurred to me that ejector seat rails on military aircraft are quite standardized, so you could replace it by a rack of hardware that sits in the cockpit and (fairly easily) transfer it to other aircraft afterward.
Unfortunately apart from converting a combat aircraft into a drone relatively easily I cannot think of any other uses.
Quick testing of a new kind of sensor (especially if it needs to see the sky) without having to fit it into a pod? Some sort of sat comms ?
All sounds a bit far fetched. A neat hack but not actually that useful.
Anyone else have any ideas?
-
Wednesday 17th May 2017 21:30 GMT SkippyBing
Re: All of this has happened before, and all of this will happen again
'All sounds a bit far fetched. A neat hack but not actually that useful.'
I think it was mainly as you say to convert combat aircraft into drones. Although the UK programme never really got anyway beyond a couple of Sea Vixens there were aspirations to convert a lot more and a number of retired Lightnings as targets. The USN and USAF have done this and converted a few hundred F-4 Phantoms to drones for missile test purposes, and they've now moved on to high mileage F-16s, so some sort of UDP may have sold to them although I don't think that's the route they've taken.
-
Thursday 18th May 2017 08:35 GMT John Smith 19
Re: All of this has happened before, and all of this will happen again
"I think it was mainly as you say to convert combat aircraft into drones. Although the UK programme never really got anyway beyond a couple of Sea Vixens there were aspirations to convert a lot more and a number of retired Lightnings as targets. The USN and USAF have done this and converted a few hundred F-4 Phantoms to drones for missile test purposes, and they've now moved on to high mileage F-16s, so some sort of UDP may have sold to them although I don't think that's the route they've taken."
I do love the convenience of the concept. It's just so neat. It put me in mind of the "Q hatch" in the U2. Just a rectangular tunnel running top to bottom behind the pilot you could insert whatever you wanted (usually, but not always a camera package of some kind, with one looking at the stars to get a precise location) into. In principal the cockpit environment is a bit more hospitable in terms of temperature and humidity than inside a wing mounted pod and the visibility to the sky is much better, hence my thoughts of sat comms.
Oh well.
-
-
-
-
Wednesday 17th May 2017 09:11 GMT Milton
BS detector activated
Sorry, I have to call BS on this, and I don't suppose I'm the only one. The physical bot-in-the-seat (let's call him Otto?) must have at least as many inputs available as would any built-in system. It must be able to make at least as many outputs (control actions) as a human pilot, but why would you limit it to human-only capability, when built-in software can do so much more all at once? And Otto's software cannot be less complex than a built-in system: it has to be more complex because it requires extra code for running otherwise pointless servos and pressure sensors and wotnot. Plus, if it even needs to be said, why waste an entire seat for machinery, when you can have built-in systems and two seats available for humans?
And let's not enquire what happens when Otto sees—if he even can see—a runway incursion ahead, where a human pilot would hit TOGA as a brainstem reflex?
So logically this makes sense only as a "dare", per the headline, with perhaps some peripheral learning opportunities. But no one with half a brain would create and install Otto as a right-seater when you can build in better (and cheaper, I'd guess) embedded systems. Perhaps the real exploration is looking into general-purpose robotic systems capable of replacing humans for certain highly rules-based tasks? In that context this fanciful test might make a little more sense.
Unless, of course, the F-35 program is in such dire trouble that ancient, pre-software planes are going to be dusted off for service? It'll be something to see: squadrons of alpha-version terminators clanking as they board old century-series fighters in the desert ...
-
Wednesday 17th May 2017 12:29 GMT Alien8n
Easy to fly
The 737 is actually remarkably easy to fly, with even minimal training. Had the pleasure of several hours in a 737 simulator at Gatwick prior to it being shipped off to Seattle, most of it is automated but I was surprised by just how easy it was to land. That said that was under perfect conditions, not sure I'd be quite so successful given more adverse weather conditions.
-
Wednesday 17th May 2017 17:03 GMT EveryTime
It's DARPA -- they have earned the right to do 'pointless' research
My first reaction was the same as others -- "why not just include the functionality in the avionics?"
Then I thought about why so many private planes had only half century old avionics installed, and a Garmin GPS temporarily mounted. The GPS was very expensive compared to identical spec units used for driving, but still vastly less expensive than properly installing and integrating approved avionics. Plus the typical pilot can easily buy a just-released far better unit every few years and leave the older GPS around as a back-up.
-
Wednesday 17th May 2017 19:48 GMT Anonymous Coward
Nice job
As a proof of concept, it's pretty cool. Still has challenges, but DARPA is at least if not more about the art of the possible as appearing-now-in-stores product.
People who fly commercial planes think of themselves as having special skills (skills that justify lots of training and largish salaries). I don't take a position on that. I do predict that in a few years time there will be an absolutely brutal fight between commercial pilots and airlines when the latter try to introduce this technology into regular flights. People kind of write off working class occupations displaced by technology (no one cares about truckers being automated away) but middle-class jobs elicit howls of anguish. Steel yourself.
It certainly doesn't help that since 9/11 commercial pilots barely interact with the passengers any more, and the flying experience is more like riding a bus every day. In that environment, the public may conclude that an automaton or a remote pilot can drive the bus just as well as Captain Yeager up there behind the magic steel door.
-
Wednesday 17th May 2017 20:00 GMT Anonymous Coward
Re: Nice job
But then what happens when crap happens such as those occasional bouts of severe turbulence, an urgency like Qantas Flight 30 (when the air cylinder blew and cause a depressurization), or even an emergency like the multiple bird strikes that forced down US Airways Flight 1549 and took a very quick-thinking pilot to think about ditching in the Hudson River?
IOW, for at least the forseeable future, you can probably take ONE pilot out of the cockpit but not both because machines still have trouble dealing with Murphy.
-
Thursday 18th May 2017 01:56 GMT Anonymous Coward
Re: Nice job
If the airlines want it badly enough, they can make it happen. And the airlines *hate* the pilots (more accurately, the pilots' union). Big salaries, publically vocal, nice pensions, strict limits on hours, it's enough to make your average Airline Wall Street CEO quite grumpy.
I can see autonomous systems aiding the pilot/co-pilot first, then the co-pilot goes and is replaced by automation with ground based pilot backup (like a drone operator) and then eventually the pilot is either gone or converted into a glorified bus driver. I don't particularly welcome this scenario (for the reasons you cite) but airlines will do anything to save money, and pilots are one of the few highly-paid, highly skilled unionized workforces out there, and certain political classes would just love to break that one up.
-
-
-
Friday 19th May 2017 14:31 GMT baron
First congratulations to the programmers for this achievement.
Now, when I read the headline, I figured the robot was actually flying the plane by hand. It was using the autopilot. Big deal. So it spun an airspeed dial and pulled the throttles back to reverse. The plane already has an auto-land and used it to land the plane. That is where the real work is.
I will say, its going to be a real long time before the general public will take an airliner with 160 passengers without two pilots. Two pilots are their to make sure that each decision that is made is a good decision. One brain fart, or bad decision could kill a lot of people. Look at how many times planes have landed at the wrong airport. And, that's with two pilots!
Andy