Flight Simulators

[Guest]

Only because the list of crashes that would've been caused by computer errors, but averted by human intervention, is not public knowledge - they exist only as internal safety memos - for corporate image reasons. For example - the amount of errors that comes up everytime an airline navigation database gets updated (every 28 days) is often downright scary. Our most recent database update had an error that "crashes"(ie fails) both flight management computers entirely every time a particular approach at a certain airport was loaded - which means losing most navigation and automation functions instantly. Other times an error is more insidous - ie a wrong altitude - but the results can be just as deadly without a human preforming a "sanity" check in realtime.

NAV update causes FMS crash? Very funny. I hope manufacturer is already gone bust?

There is a difference of sanity checks that can be formalized and automated, and those that cannot at our current AI knowledge level. But sooner or later, we will have enough knowledge and knowhow to automate ANY decision making. AI can be made perfect, humans are always faulty at some point.

[Columbia]

I'd gladly see pilots replaced with AI. List of perfectly healthy aircraft crashed due human error, bad judgment and bravado is a lot longer, that list of damaged aircraft saved by human creativity and non-standard thinking. And AI don't need training - it's proficient out of the box.

If people were replaced by robots, many would lose their jobs, perhaps affecting economy.

I wouldn't be surprised if there would be people protesting against computer-driven trains.

[AngelLestat]

That looks awesome, of course oculus rift does not remplace that, but it may be a most cheaper choice for pilot students to begin and to increase its experience...

Because they can not be all day inside that.. there will be many more in the line waiting.

[Neil1993]

An Oculus Rift type of device could have a place in training.

If you are running a flight school or something similar with 200 plus students, you can't put one or two students in a big simulator for 20 some hours a week.

You also can't buy too many of the big simulators at 10 to 20 million ether, but you can supplement their training with an Oculus Rift like device.

A student can get the basics down before going into the full simulator.

You'd be surprised how many simulators some companies buy. Some of the bigger players could buy up to 10 at a time, and I know my company ships anywhere between 20 and 50 simulators a year.

Only because the list of crashes that would've been caused by computer errors, but averted by human intervention, is not public knowledge - they exist only as internal safety memos - for corporate image reasons. For example - the amount of errors that comes up everytime an airline navigation database gets updated (every 28 days) is often downright scary. Our most recent database update had an error that "crashes"(ie fails) both flight management computers entirely every time a particular approach at a certain airport was loaded - which means losing most navigation and automation functions instantly. Other times an error is more insidous - ie a wrong altitude - but the results can be just as deadly without a human preforming a "sanity" check in realtime.

Wildlynx is right. Current computer technology might be a long way from perfect (and it seems like your FMS is even further) but computers and AI have the advantage that they do not have to be re-trained constantly, the improvement that you give one can be given to any other and they retain any improvements made (if you want to replace a human with advanced skills, you have to train another human).

[cpast]

NAV update causes FMS crash? Very funny. I hope manufacturer is already gone bust?

There is a difference of sanity checks that can be formalized and automated, and those that cannot at our current AI knowledge level. But sooner or later, we will have enough knowledge and knowhow to automate ANY decision making. AI can be made perfect, humans are always faulty at some point.

So your idea is that while humans are always faulty, humans can make perfect AIs? Yeah, that's not happening. Ever. There will never be a bug-free AI. All you can get is an AI with no *known* bugs that makes better decisions in general than a human; you cannot have a perfect AI. I view the claim that we'll be able to automate all decision making (and that this will be a good idea) as unlikely in any but the very, very, very far future, but it's possible; a perfect AI is not possible.

[mrfox]

NAV update causes FMS crash? Very funny. I hope manufacturer is already gone bust?

EFF 09 DEC 14:

CAUTION: 747 FM DATABASE RNAV STARS KARBU 1N/GUKDO 1N/OLMEN 1N/BIKSI 1M EXCEED STAR DESIGN MAX WPTS DUE MULTIPLE ALT/SPD CONSTRAINTS AND MAY CAUSE DUAL FM FAIL. DO NOT ACCEPT THESE RNAV STARS. OTHER USEABLE STARS GUKDO 1A AND MAKSA 1M. KOREA CAA INFORMED BY BOEING AND AIRLINES. STAR REDESIGN PENDING.

Not funny at all when the crew found out for the first time on approach. Don't see Boeing going out of business over this anytime soon either.

[WestAir]

The arrival was so complex it crashed the FMC?

What airport was that? I need to see that star.

[mrfox]

Current computer technology might be a long way from perfect (and it seems like your FMS is even further) but computers and AI have the advantage that they do not have to be re-trained constantly, the improvement that you give one can be given to any other and they retain any improvements made (if you want to replace a human with advanced skills, you have to train another human).

From what I understand, is that with current technology, a computer still cannot be aware of a problem that it was not programmed to diagnose, or "improvise" a new solution to a problem that has not been considered in its design. What the future may hold? Well that's in the realm of speculation. Would an AI that has the capacity to "foresee" a situation with a sub-optimal outcome for the humans - be required know and understand the same "motivations" humans have in the first place?

- - - Updated - - -

The arrival was so complex it crashed the FMC?

What airport was that? I need to see that star.

Not overly complex, just designed in a "dumb" way to have altitude and speed constraints at almost every waypoint - a situation that was not considered when the 747 FM was designed.

Airport is Seoul Incheon - ICN/RKSI.

[linkxsc]

Well my humble opinion.

I do a lot of joystick games. (Elite Dangerous, War Thunder, Il2, LockOnMC, DCS to name a few) and in the case of elite and war thunder I have had opportunity to play with an Occulus Rift, and track IR (Im poor though so im usually restricted to a hat switch for view controls)

Now dont get me wrong. OR and trackIR can VASTLY improve gaming experience and situational awareness when compared to a hat switch. But they would never be able to match gaming in a cockpit with 360 degree vision.

Ofcourse this is just a gamers perspective of those systems. The second you start trying to train pilots though, those full cockpit simulators are probably a must next to actually flying.

[rpayne88]

Stupid question: Does Canada have the same 20 hour max rule for pilot training. As in, you can fly as many hours as you want in it, but only 20 will count.

(I think its 20 hours max.)

[Sandworm]

Imho the occulus will overcome much of these simulators, probably reducing their market. While no sane person would say they provide an equivalent experience, there are things that could be offloaded from expensive simulators to the, now, more capable desktop thanks to the occulus.

Once upon a time they build massive simulators that used real terrain, miniatures with a camera/scope on a pole "flying" through them. These were massively expensive but did allow experiences (fog/reflections/sunlight) that no electronic sim could hope to match until very very recently. At the time everyone laughed at computer-based sims. But while the computers of the time could not provide an equivalent experience, they could do some things very very well (ie IFR navigation). So some of the training moved from the expensive simulator to the desktop. The same will happen today. Some part of the expensive simulator's role will shift to the cheaper but increasingly capable occulus. I'm not sure exactly which training will move first, but am in no doubt that something will.

See https://design.osu.edu/carlson/history/lesson13.html

And http://en.wikipedia.org/wiki/File:TL39_Flight_Simulator_Visual_System.jpg

[Neil1993]

Imho the occulus will overcome much of these simulators, probably reducing their market. While no sane person would say they provide an equivalent experience, there are things that could be offloaded from expensive simulators to the, now, more capable desktop thanks to the occulus.

Again, the problem with the occulus, which I've already pointed out, is that it's unlikely to ever be qualified as one-to-one flight experience. By one-to-one flight experience, I mean one simulator hour counting as one hour in an actual aircraft. Pilots still need these hours before they're allowed in real commercial jets.

[DaveofDefeat]

I would love to see somebody try to press buttons while wearing the oculus rift.

Edited April 11, 2015 by DaveofDefeat

[Sampa]

Op, i want your job!

[Astronut]

For training I seriously doubt that a professional grade sim will be replaced any time soon. There is a lot of value in training in a true copy of the cockpit. In fact, that has far greater importance than fancy visuals.

All the simulators I have had the opportunity to train in have had inferior visuals when compared to current desktop computer entertainment sims. But then most of the scenarios seem to be at night in solid IMC where you are too busy dealing with simulated failures to appreciate the view out of the windshield.

VR headsets will certainly have a place in pilot training though IMHO. I think they would be great as a cockpit familiarization tool. Also for simulating tactical scenarios where simulation of the airplane isn't the prime purpose of the training.

[Kerbart]

From what I understand, is that with current technology, a computer still cannot be aware of a problem that it was not programmed to diagnose, or "improvise" a new solution to a problem that has not been considered in its design. What the future may hold? Well that's in the realm of speculation. Would an AI that has the capacity to "foresee" a situation with a sub-optimal outcome for the humans - be required know and understand the same "motivations" humans have in the first place?

At the same time, and I'll admit this may be a cynical or cold-hearted approach, if this means one crash per x-flights replacing five crashes per x-flights due to pilot error, you'd still be ahead of the game. If human pilots were behaving perfectly we wouldn't be having this discussion either.

As much as we think technology advances by revolution, and in the grand scheme of things it does, the actual process is always, always evolutionary. My prediction is that you will see certain airports that will become certified for robotic flight, and there will be very stringent rules for that, not just for the airports but also for the airplanes and the robots involved (mechjeb need not apply). But there's advantages for both as you can probably squeeze more slots in a day. Over time you'll see more airports and more airplanes using robotic flight, and large airports like Atlanta, LAX, JFK, etc will probably become robot-only airports. And 100 years from now we'll be wondering why there was such a resistance against it, just like we laugh now over the concerns of the early steam trains that would go so fast that passengers would pass out because of the lack of air at those speeds.

[mrfox]

At the same time, and I'll admit this may be a cynical or cold-hearted approach, if this means one crash per x-flights replacing five crashes per x-flights due to pilot error, you'd still be ahead of the game. If human pilots were behaving perfectly we wouldn't be having this discussion either.

As much as we think technology advances by revolution, and in the grand scheme of things it does, the actual process is always, always evolutionary. My prediction is that you will see certain airports that will become certified for robotic flight, and there will be very stringent rules for that, not just for the airports but also for the airplanes and the robots involved (mechjeb need not apply). But there's advantages for both as you can probably squeeze more slots in a day. Over time you'll see more airports and more airplanes using robotic flight, and large airports like Atlanta, LAX, JFK, etc will probably become robot-only airports. And 100 years from now we'll be wondering why there was such a resistance against it, just like we laugh now over the concerns of the early steam trains that would go so fast that passengers would pass out because of the lack of air at those speeds.

I have no doubts that the evolution of automation in transportation will continue towards full automation. My opinion is only that the readiness and reliability of the technologies required are quite a bit beyond the current state of the art, and it will be some time before such a reality can be realized.

One of the big issues, as you correctly identified, is the mixing of automated traffic with human controlled vehicles. Air traffic is by nature free flowing, analogous to vehicular and ship traffic, and crisscross from branch to branch. Unlike an automated subway line, where you can built a isolated system that is segregated from non-automated traffic, it would be extremely difficult, and operationally and economically unfeasible, to set aside space for automated aircraft where they won't interfere with human controlled traffic.

Yes, google is testing driverless cars, but a human driver is always behind the wheel, ready to take over at a moments notice. So really it is actually only catching up to the level of automation that already exist in current airliners today during low visibility operations.

certain airports that will become certified for robotic flight, and there will be very stringent rules for that, not just for the airports but also for the airplanes and the robots involved (mechjeb need not apply). But there's advantages for both as you can probably squeeze more slots in a day.

The reality today is that when airports today go into low visibility operations and automatic approaches and landings are being conducted, the operations rate dramatically reduces to cater for the increased separation requirements.

[Sandworm]

Yes, google is testing driverless cars, but a human driver is always behind the wheel, ready to take over at a moments notice. So really it is actually only catching up to the level of automation that already exist in current airliners today during low visibility operations.

Actually, they aren't heading that way. Google is aiming for cars without human controls, and have been slammed for it by regulators a couple times. (see http://www.wired.com/2014/12/google-self-driving-car-prototype-2/). It boils down to market forces. Google doesn't want to compete with Ford and Toyota for drivers. Google wants to hit the untapped "grey" market of people who do not today drive. The over 65s have most all the money in this world and will pay anything to keep their "mobility". Google wants to give them cars for which you do not need a license. The scary end goal is that they want to give cars to the legally blind.

Edited April 11, 2015 by Sandworm

[mrfox]

Its not where they are headed - but its where they are now. The step over to full autonomus control with no realtime human backup - and no human to take responsibility immediately when the automation goes wrong, will be a huge one. One of the biggest challenges is whether a system is fail-safe - an automated subway system can simple put the brakes on and stop, if it falls outside what the system consider "normal" parameters, and be confident it will safely come to a stop without many issues, because it is on rails. And because all automated train systems today are segerated systems, they wont be putting any other traffic - train or human, at risk either. To be able to prove statistically that a driverless car, at speed and in mixed traffic, can "always" provide a safe outcome for everyone if anything goes wrong at speed, is a challenge and responsibility that current technology is not quite ready for, and the reason they get "slammed" by the regulators.

Airplanes, even more so, as they go much faster and can't stop in midair.

Edited April 11, 2015 by mrfox

[Sandworm]

Its not where they are headed - but its where they are now. The step over to full autonomus control with no realtime human backup - and no human to take responsibility immediately when the automation goes wrong, will be a huge one. One of the biggest challenges is whether a system is fail-safe - an automated subway system can simple put the brakes on and stop, if it falls outside what the system consider "normal" parameters, and be confident it will safely come to a stop without many issues, because it is on rails. And because all automated train systems today are segerated systems, they wont be putting any other traffic - train or human, at risk either. To be able to prove statistically that a driverless car, at speed and in mixed traffic, can "always" provide a safe outcome for everyone if anything goes wrong at speed, is a challenge and responsibility that current technology is not quite ready for, and the reason they get "slammed" by the regulators.

Airplanes, even more so, as they go much faster and can't stop in midair.

I never said it was a good idea, just that that's the way they see the market. I don't think they have to make a car absolutely safe, just better than the minimum standard for drivers today. Given that machines don't drink, race, speed or have strokes, it might not be all that difficult.

Airplanes, even less so imho. An airplane flies in a rather homogeneous (air) and regulated (FAA) environment. Even today probably 90% of "heavy metal" landings are done by machines. If you go by hours, probably 99.99% of flying hours are in the hands of autopilots already. Automated cargo craft are probably doable today.

[cpast]

I never said it was a good idea, just that that's the way they see the market. I don't think they have to make a car absolutely safe, just better than the minimum standard for drivers today. Given that machines don't drink, race, speed or have strokes, it might not be all that difficult.

Pilots are much, much, much better-trained than drivers. Pilots of airliners have extensive experience. Beating the average pilot is orders of magnitude harder than beating the average driver.

Airplanes, even less so imho. An airplane flies in a rather homogeneous (air) and regulated (FAA) environment. Even today probably 90% of "heavy metal" landings are done by machines. If you go by hours, probably 99.99% of flying hours are in the hands of autopilots already. Automated cargo craft are probably doable today.

Most landings are not autoland, because autoland is higher-workload than normal landings. Autopilots cannot fly a plane autonomously; they need monitoring, and don't even try dealing with abnormal situations (they disconnect if things start going wrong). That's part of how they can be so reliable in normal operation -- they simply don't control the aircraft if things have started to go wrong. A car has an easy stable safe mode. A plane, no.

[mrfox]

I never said it was a good idea, just that that's the way they see the market. I don't think they have to make a car absolutely safe, just better than the minimum standard for drivers today. Given that machines don't drink, race, speed or have strokes, it might not be all that difficult.

The biggest difficulty will be interacting with other human vehicles and pedestrians.

Airplanes, even less so imho. An airplane flies in a rather homogeneous (air) and regulated (FAA) environment. Even today probably 90% of "heavy metal" landings are done by machines. If you go by hours, probably 99.99% of flying hours are in the hands of autopilots already. Automated cargo craft are probably doable today.

Autolands are rarely done - not only because of the higher workload, but because greater tolerances are required - the sensitive approach area must be kept completely clear of traffic, and the seperation between aircraft greatly increased over visual operations. The handling capacity of the airfield dramatically decreases. Autolands are generally only done when visibility precludes normal visual landings, and for crew/aircraft certification/currency.

[linkxsc]

I would love to see somebody try to press buttons while wearing the oculus rift.

Tried it. I guess its doable. But for gaming, you basically need a gamepad controller or a HOTAS setup so that all controls are at your fingertips. Cant really go hunting around a keyboard for the flaps key while dogfighting with an Oculus Rift

[mrfox]

Where 3D goggles are really going to shine is in remote operated, camera equipped drones (UAV) operations. A huge and common problem for drone operators today is the total loss of situational awareness - often the phrase "where am I? And what did I just hit?" come to mind. If fact, if the drones mission is survalliance/reconnaissance, it would provide the most natural user interface if combined with a hovering/rotary wing craft, and control algorithms to emulate a "mouse look" environment.

[noobsrtoast]

As far as professional training simulators go, I don't think an Oculus Rift provides the same experience as a real cockpit, with real seats, real instruments, real switches, and real person sitting in the seat next to you.

it never could, simulations are just that...simulations