Automating the Air

How will self-driving cars affect us? Automation has been in the cockpit of airplanes for decades, how has it worked out for the pilots who fly them?

I was discussing self-driving cars with my friend Matt, who is a pilot for a major US airline. You have definitely heard of it, and there is a very good chance that you’ve flown it. Matt pointed out that planes have had automation for decades: the trusty autopilot feature that gives Tesla’s AutoPilot its name. I realized that the issues that drivers are struggling with now — such as how much to trust the automated system — are ones that must have been encountered by commercial pilots for some time.

To explore this, I set up an interview with Matt about automation in the cockpit. Here’s the transcript:

Loren: Okay. You’re a commercial pilot, Matt, for a -

Matt: A major U.S. airline.

Loren: When someone says, “Which one?” I say, “A major one.”

Matt: Yeah. Major. There’s only three, so … Three and a half if you count Southwest. A major U.S. airline. You know my background. 16 years at a regional carrier, now with a major carrier, so I have some perspective about automation and the way we divide tasks in the cockpit between doing it manually or letting automation do it, and then checking and confirming, monitoring systems, overriding systems when they’re not doing what we want them to do, or not. Letting them do what they’re designed to do and just hoping we have them set up correctly. It’s about managing the systems.

Loren: What has changed in 16 years? How are things different now than when you started?

Matt: I don’t know that they are. When I first got hired I flew a turboprop propeller airplane down in the Caribbean, then I transitioned off that to a jet, and I flew that for 15 years. It’s not like I was bouncing around from airplane to airplane, encountering a bunch of different designs. I was flying the same airplane for 15 years.

I’m now on a new airplane by the same manufacturer, so it is an updated design from the one I was flying for 15 years. I can talk a little bit about some of the different philosophies or maybe some of the lessons learned by the designer.

Loren: What sort of stuff gets automated on an airplane?

Matt: There’s this idea of that “autopilot” means we just push a button and it flies itself. That’s not accurate.

There’s two ways to fly the airplane. You have the yoke, which is the steering wheel, and the throttles, which control the power. It’s equivalent to the gas pedal in a car, and you can fly the airplane that way, and that works fine. Your partner will set courses and altitudes and headings and you will fly to those parameters.

Loren: They’re like a navigator, basically.

Matt: Yeah, basically. Tell them, “Hey, I want to fly a heading 3, 3–20 heading.” He sets the little heading bug it’s a little pointer and you turn the airplane to fly to 3–20. That works fine. Or you can autopilot on, and then if you set the heading bug the autopilot will just manipulate the control services of the airplane in order to turn the airplane to that heading, or that altitude, or whatever. There’s varying levels of automation.

Loren: What was the original reason for autopilot? Why did they bring it in in the first place?

Matt: Task loading. If you’re just sitting there watching the autopilot fly the airplane, you’re not as task loaded. Especially if you’re on a takeoff, climb-out, or on the descent approach landing, it’s a high-task environment, and the idea was it takes some of the workload off the pilot.

Also, it dates back to when it was a three-man cockpit. You had the captain, first officer, and flight engineer. The flight engineer would be manipulating all the systems, the hydraulics, the air, all that stuff. The two guys in front would be concentrating on flying. At some point, this system became automated enough, you didn’t need that third guy anymore.

Loren: That’s the first job lost, right there, basically.

Matt: Yeah, and that was actually an issue for the unions and collective bargaining and stuff, “Oh shit, that’s 33% of our jobs are now gone.” That was the DC-9. That was the first two man airplane. The first big, jet two man airplane. The idea was, well, gosh, we really need to, now that it’s only two guys instead of three, we really need to take a lot of this task loading off of the only two remaining pilots.

It’s not just autopilot, it’s a lot of the systems will automate themselves, like this valve will open, this valve will close, automatically, in reaction to certain situations. The fuel system, the air system, the hydraulics, the valves are opening and closing, whereas before the flight engineer used to manually open and close all that stuff. That’s automation as well.

Loren: When that stuff is happening automatically, do you need to understand all of what’s going to happen in your head, or is that stuff that’s like, “It’s a black box, we’ll just let it take care of itself.”

Matt: It’s mostly a black box. If there’s an error, you get an error message. There’s a screen on the front and center of the panel that’s got all the error messages, and there’s hundreds of different error messages, and if you get one you look it up in the book and do the procedure. If you need to, then manually override or something like that, then the checklist will tell you how to do that.

In training you should have a basic — you don’t have to build the airplane. You’re not going to build it or repair it, which some training programs do make you build the whole damn airplane, but you should have a fundamental, rudimentary understanding of the systems so that when things are going haywire you can know what’s going on, or understand why it’s doing what it’s doing or more importantly what it’s about to do. How one error message or one failure is going to lead to these other failures and you need to be prepared for that.

There’s a checklist for every little error message that we get. There’s a procedure, and we’re trained not to think too much, and just open up the book and do the procedure.

There are a few that are time critical and those are memory items that we do have to have memorized and just do immediately, like if there’s a rapid cabin depressurization, and you’ve only got a couple of seconds of useful consciousness, you got to get the oxygen mask on, you got to establish the intercom, you know. There’s some time-critical things like that.

Loren: Most of the systems we’re talking about that are automated, they have to do with the flying of the plane, right?

Matt: Well, yeah. Pressureization, pneumatics, hydraulics, fuel, electrical. A lot of electrical. Airplanes are, because it’s so automated, there’s a lot of electricity flowing around airplanes. There’s redundance, and every system has at least one, and usually two or three redundant systems. Those are the kind of things that I’m talking about, that are sort of black box-y. The anti-ice which is also part of the air and pneumatic system. Yeah. Fuel.

Loren: I’m used to the world of crappy commercial software. One of the experiences that I’ve had is that new things are brought in, and they go through a period of unreliability, then eventually they mature, they become things you can count on. Does that happen on airplanes, as well?

Matt: Definitely. Just like what happened in cars, they say don’t buy the first model year of a car. There’s only two major manufacturers of commercial airplanes, there’s two more manufacturers of smaller commercial airplanes, the regional jets. They’ve been doing it long enough now that a lot of those bugs are ironed out. Obviously, when Boeing was making the 787, was rolling it out a couple years ago, that was a radical redesign because they’re using all these composite parts, they went through a different manufacturing process where the supply lines were worldwide and different components were being manufactured all over the globe, so when one component wasn’t ready or wasn’t working for some reason, that messed up the whole development and rollout of the airplane. I think that’s a one-off.

Loren: But that’s a supply chain problem. That’s not the things going wrong in the air, right?

Matt: No, it led to some things going wrong in the air. When they first wheeled it out, the batteries were overheating, which can be a serious problem as we’re learning about with these Galaxy 7 phones.

Loren: You had talked a little bit about the difference in philosophies of the different airplane manufacturers, and how abstract they are in terms of moving the experience of flying an automated plane away from emulating a manual flight. You want to tell me a little bit about that?

Matt: The two major manufacturers are obviously Airbus and Boeing, Boeing being the American company, Airbus being the European company. They’re kind of a consortium. You have two different design philosophies to deal with … Getting away from what we were talking about as far as the systems and valves are opening and closing and stuff, now I’m talking about actually manipulating the airplane and maneuvering it. Airbus, they call it set it and forget it. There isn’t the old-school control wheel yoke. There’s a little joystick off to the side. That’s it.

Loren: Is that weird for pilots?

Matt: Yeah, they say it takes a little getting used to. A couple of days in the simulator, they say, and then you’re fine. It’s definitely not something for those of us that have ever flown an Airbus product, that’s not what we’re used to.

Then the thrust levers, that is, the throttles. They’re usually on the center pedestal, in between the pilots. When it’s time to take off, you just bring them up and you set them in the notch, and that’s it. In the Boeing products, you set them in the notch, but then as you are climbing up, leveling off, and descending, there’s a little servo motor in there that actually moves those thrust levers back and forth to simulate what you would be doing if you were flying it by hand.

Loren: How does it work manually, if it was just a manual throttle what would happen?

Matt: It’s just a rheostat, like a dimmer switch. You just move it back and forth.

Takeoff, that’s the most thrust because you need all your power to up get off the ground, in cruise it would be less thrust, and then descent it would be even less thrust. Just like going up a hill or down a hill in your car, you have to press down on the gas pedal to go up the hill, you have to ease off the gas pedal to go down the hill. Exact same.

Loren: The Boeing thing is, it is making those movements to suggest that stuff, and in the Airbus product, it sits where it sits.

Matt: It sits where it sits. You have dials on the panel that shows you how much power — with the gauge — that the engine is giving you. How fast its turning, all that kind of stuff. That’s true on both airplanes. It shows visually how much power the engine is putting out, but the Boeing, the American Boeing model simulates what you would be doing if you were manipulating the control, and the same with the yoke. It’s got little motors that will actually turn the yoke to simulate what you would be doing if you were hand-flying.

Loren: What’s the impact of that. Is it just aesthetics, or difference in pilot error?

Matt: No. In small airplanes or in really old airplanes there were literally a system of cables and pulleys that went from the control wheel to the control surfaces. The rudder and whatever. That’s gone. That’s not the case anymore, it’s just a rheostat that activates a hydraulic actuator.

The idea is to keep the pilot “tactile-y” connected to what’s going on, whereas Airbus just eschews that whole thing and just gives you a visual indication on the screen about what it’s going the joystick just stays where it’s at with the autopilot on the thrust levers just stay where they’re at and you just have indications on the screen telling you what it’s doing.

I don’t think either is better, I just think it’s a different philosophy.

Loren: There’s not an impact? They haven’t looked at one and said, “This is safer because pilots won’t take it the wrong way or they won’t overcompensate for something, or they won’t …”

Matt: No, because, the logic in the computers doesn’t allow you to over-compensate.

Loren: Well that’s interesting.

Matt: Yeah, it limits what you can do on both products. On both products, it’ll limit certain things and not let you do something stupid anyway.

I’ll give you an example. In aviation, the term stall means something different than it means to drivers. In driving, to stall your car means the engine crapped out and now you’re just dead on the side of the road. In aviation, a stall is when there’s not enough air going over the wings to maintain flight.

You can recover from it. Airplanes are designed such that there’s not enough air going over the wings so now the nose is going to drop, we’re going to build airspeed again, we’ll get air going over the wings and go flying again. It’s not the end of the world, but it’s still not good. The way a stall happens is, if you pitch up too much without enough power, now the air hitting the airplane is hitting the belly of it instead of hitting the front of wings, and that’s how you stall.

Neither product, the Boeing or the Airbus is going to let you pitch up enough to stall.

Loren: How will it stop you?

Matt: It can sense what is called relative wind, so it can sense the angle that the air is hitting the airplane, and before it reaches that critical angle, it won’t let the airplane pitch up anymore. It’ll just keep where it’s at and you can keep pulling back, pulling back, or pulling that joystick back, back, back but it’s going to limit you so that you don’t stall, because the theory is that it’s better to remain in control than not. Even if you think you want to — you’re sinking, sinking, and you’re trying to arrest that sink, normally you pull back, but if you pull back too much, now you’re going to stall.

Loren: If you stall you would basically lose control of the plane?

Matt: Correct. There are ways to recover, but it’s better not to lose control in the first place.

Loren: So, there is no scenario where it would be a better thing? I’m thinking what if you’re trying to avoid hitting something.

Matt: That’s a great example. You’re flying along at night in South America, there’s 25,000 foot mountains the computer alerts you there’s a mountain ahead of you and you go, “Oh, crap.” You go full power and now you have to pull up as much as you can. You’ve got as much power as you can get in there, you’re still trying to clear that mountaintop, you might overdo it but the logic’s not going to let you. That’s a great example of why that logic is there.

Loren: I remember seeing the cockpit when I’m entering or exiting the airplane, that kind of thing, and it does seem like it’s changed a bunch. I have memories of basically a million switches, and then in more recent times hopefully more modern planes, I’m seeing a lot of screens, with, basically it looks like a lot of software with a lot of screens on there and maybe less switches. What’s materially changed?

Matt: Here’s something interesting, the FAA, when they’re certifying airplanes, manufacturers put out different models of the same airplane, longer fuselages, shorter fuselages, but the manufacturers want to have an incentive not to update and modernize because they want to be able to call them common types with the old model, so that if an airline bought a hundred of the old model but the manufacturer now wants to put out a new model, they need to make the cockpit look the same as the old model so that the pilots and the mechanics and the flight crew can just swap back and forth between them and not have to be trained on a whole new airplane.

Loren: This slows down the rate of progress, basically.

Matt: Correct. Correct, because they have to convince the FAA that a pilot can fly one leg in this old model, get up out of the airplane, go across the terminal, fly another leg in this new model, and it’s similar enough that he doesn’t need a bunch of retraining. He can just do it, because that’s much more efficient for the operators, for the airlines.

There’s this incentive to keep the cockpit what they were so that they can keep selling new versions of the airplane, basically. The 737 is a great example of that. That airplane also has been around since the ’70s, and it’s gone through five or six different models. The panels and everything, the systems are certainly updated from the original but each iteration is only a little different, because they want to be able to tell the airlines, when Boeing’s going to try and sell the new model now, “Yeah, we’ve made these improvements, but it’s still similar enough to the old airplane that we’ve convinced the FAA that your pilot won’t need new training.”

Loren: You spelled out a vision to me at one point of how fully autonomous airplanes could come to be and an intermediate step where there’s pilots on the ground flying airplanes in the sky.

Matt: Sure. We have that, now, right? We’ve had guys on the weekend go flying remote controlled airplanes. You could do that. That technology’s been around for decades, and now obviously in the air we have drones. Unmanned airplanes. That technology is there. Obviously, there’s security issues. Obviously all the hacking going on now, I think there’s a public relations issue, like when nervous fliers get on an airplane. You don’t want to look left as you’re boarding the plane and see one or no guys in there. There’s just a thing about people like to see people in the cockpit.

The fact is pilots make mistakes just like drivers make mistakes and the automation catches a lot of those mistakes just like the automation in cars might catch a lot of mistakes. The automation will also make mistakes, just like car automation will make mistakes. Neither one is inherently better.