How Does Autopilot *Actually* Work?

Autonomous vehicles, autonomous drones, autonomous planes....??? It's hard enough to understand how planes fly, but how do they fly themselves?! Or more accurately, how do the computer systems on board, aka autopilot, control the plane? And how much are commercial pilots *actually* flying, anyway? Well, it's kinda a long story.

In this video, I take you through the history of how planes became "electrified", through first the invention of Fly By Wire by Airbus, where mechanical controls on the aircraft were replaced by computers and actuators/servomechanisms. Then, I'll discuss some of the brilliant algorithms used to filter the sensor data, handle different errors, and even land the aircraft completely on its own, also known as a CAT IIIc, or zero-zero autoland. I'll take you through the system architecture of an autopilot system and its sistem, the autothrottle! (it's late, don't judge me).

Of course! It's not a real algorithm video without mentioning Artificial Intelligence for aviation, or AI for A! I'll talk about NASA's Learn To Fly developments, and other research programs on what the future of autopilots may be..... and if computers could actually replace pilots!

✨✨✨✨ LIKE & SUBSCRIBE ✨✨✨✨

0:00 Intro
1:44 A Brief History of Autopilot
3:07 What is Fly-By-Wire?
7:09 How Autopilot Autocorrects Errors
9:53 Limitations of Autopilot
11:27 The Future of Autopilot

[_zaverus]

13:35

"literally learn on the fly"
omg I cant believe how well this line worked

[practicalshooter]

No reference to the super Sophisticated Auto Pilot from the movie Airplane (1980)?

[premsawh]

Nice video! Great explanation Jenny :)

[daegudiva]

I loved this content, a binge is coming

[garagemonkeysan]

Great video. Informative and entertaining. Mahalo for sharing! 🙂🐒

[ThomasArellano]

Amazing video! Wish I understood control theory a bit more to get how the filter/autocorrecting works in detail.

[itsevilbert]

So many complex topics touched upon, I'm shocked.
The big thing about the current generation of neural networks that is ignored is that they are fantastic at reacting to events that were in their training dataset, and events that are close to their training dataset. But for events that are far outside the dataset, used to train them, they have no known correct behaviour or proper reaction. And that is usually when bad things happen. It is nearly impossible to train a machine, or human, to expect the unexpected.

[gclinster]

Another great segment! The research done for this is superb.

It wouldn't surprise me if pilots watched Iron Eagle (the "other" Top Gun"). LOL

[martinan]

Future cockpit will contain Pilot, Autopilot and a German Shepherd. The Autopilot's job will be to fly the plane, the pilot's job will be to talk to the passengers and the German Shepherd's job will be to bite the pilot if he touches the auto pilot.

[Paradoxiumas]

Thanks for the video! Some bits are not a 100% accurate, which is probably due to lack of practical interaction with an autopilot. You will understand once you sit in a cockpit yourself!
What about your pilot licence plans btw? Go for it!

[MichaelBerthelsen]

Obviously pilots won't be replaced for a time.😅 People just don't trust machines the way they do human beings.😉 AI may fly, but people still want humans at the front, in case something goes wrong.😁

Great video, by the way!😉❤👍

[kyros]

Thanks for another great video! Heads up, your camera's focal zone is a bit short so your face is a bit out of focus.

[IARRCSim]

For integration drift of the motion sensor, shouldn't they merge with GPS when the motion sensor drifts outside the expected error for GPS?

[daegudiva]

"just a couple of hundreds of millions....easy peasy" Love that

[FuelAirSparkTime]

Looks like ace combat 7 will become real soon enough

[psidud]

>boeing allows pilots to push past performance limitations.

So what you're saying is...I can overclock a plane? Time to pour liquid nitrogen on the engines?!?!?

That aside, this stuff reminds me of courses I had to take in control systems and also sensors and measurement.


I think that the learn to fly stuff sounds really promising, especially if it can be done using small models of the plane instead of the full thing. If I were some director in a plane making company, I'd want to see if the learn to fly system can be used to feed back to the design of an airplane by another AI.

Human designs a "rough" draft of a plane -> AI learns to fly it -> feedback of learn to fly is used to refine the original design -> AI learns to fly -> repeat until convergence.

[jimmydore]

hmmm very interesting ! i can programm microcontrollers and now i try fpga, i didn't think on using ai for a plane, i will talk with my colleagues from systems

[networkgeekstuff]

I am from a family that has a history with rail (sorry, not aviation here). And automatic control as technology for railways was here for several decades now, but it will never replace that one person driving the train for two reasons.
1) No autopilot has general inteligence (yet) that can react to black-swan events
2) Several studies and pools confimred to the rail companies that people will NOT like to ride a train without that one person driving it, who put's his own skin on the table for that ride.

[pingnick]

Garmin auto land episode!?🤖🛫🛩🛬🗽🤯☮️💟♾🌈🚀

[telkoehf]

Lockheed L-1011 This Aircraft did not need pilat