Human versus Autopilot

Jaybird180

Jaybird180

Final Approach
Joined
Dec 28, 2010
Messages
9,034
Location
Near DC
Display Name
Display name:
Jaybird180
I know that there are several of you that enjoy hand flying and we all get enjoyment from the experience.

Yesterday, I had the pleasure of flying behind a Garmin Perspective SR22 and it is very nice indeed. We turned on the AP approx 200 AGL during the climb and played with several modes, display options, data feeds, (Warning Will Robinson - INFORMATION OVERLOAD!) etc :LOL:.

The SR22 is a very stable ship, but the AP added a measure of finesse to the manipulation of the controls. With the Garmin Perspective, there was enough information to make intelligent, well informed decisions 10 minutes ahead of the airplane (ie descending under Bravo balanced against preventing an ear busting vertical descent rate); the types of things that very experience PILOTS do on a subconscious level.

Years ago, IBM set out to prove that a computer can outthink a human in Chess, which spawned the design of Deep Blue. It wasn't successful initially, but afterward it was shown that computers can outplay human GrandMasters (Note: this claim is still disputed). Do you think that APs have evolved to where they outfly an experienced aviator? Would a human be able to match the precision and finesse of AP flying?
 
Jaybird180 said:
Would a human be able to match the precision (1) and finesse (2) of AP flying?
Click to expand...

1. No
2. Yes

The actualy FLYING part a computer will beat you everytime, but it's the easy part.

However where the pilot comes in is to make decisions and to bring the plane down safe after some kind of failure.
 
From my "All About Autopilots" presentation:

An autopilot is a cross between a zombie and a CFI.
An autopilot is like a CFI because it can fly the airplane very well, probably better than you.
An autopilot is like a zombie because it needs brains. YOUR brains. B...R...A...I...N...S...

Seriously, an autopilot will maintain your 4000 foot IFR altitude perfectly, +/- 10 feet or less, right into the side of a 6000-foot mountain.
 
TMetzinger said:
From my "All About Autopilots" presentation:

An autopilot is a cross between a zombie and a CFI.
An autopilot is like a CFI because it can fly the airplane very well, probably better than you.
An autopilot is like a zombie because it needs brains. YOUR brains. B...R...A...I...N...S...
Click to expand...

Like!

EDIT

like anything else, garbage in = garbage out
 
Last edited:
It depends on the autopilot and how it is set up by the human. I would say that generally an autopilot will hold heading and altitude better than a human. When it comes to intercepting and tracking there are large variations among autopilots due to either their software, hardware, or how they are adjusted. Some have the annoying characteristics of chasing an airspeed with obvious changes in pitch, just like a human pilot might, or s-turning down an ILS. All in all though I think autopilots are very useful in lessening the workload. The pilot does need to manipulate it correctly though. Garbage in, garbage out.
 
An autopilot will save you fuel because it can hold a heading (in calm air) way better than you can. The time you spend zigging and zagging (even very small zig-zags) trying to hold a heading adds up.

I learned this in the boating world.. in smooth seas the autopilot is better at holding a heading. When it gets rough the autopilot can't see what's coming and its better to have an active helmsman. Same goes for airplanes except maybe the ones with high tech autopilots that have yaw dampers etc... When turbulence gets past the light chop phase autopilot kinda sucks because it does not involve the rudder.
 
There's more than moving pieces around on a chess board than just making the right moves toward winning the game. No human will compete with 32 bit or even 8 bit precision accuracy that can see all the possible moves in a limited variable environment. Humans will most definitely will beat the computer in finess especially in a dynamic environment with a near infinite number of constantly changing variables.

A nice weather or predictable conditions flight from A to B like airliners do, sure, you'll have an impossible time beating the 16 or 32 bit precision of a computer if flying a specified perfect path in 3D at maximum efficiency.

Bouncy winds through a twisting mountain pass below lowish clouds then landing on an unmapped heavily obstructed narrow short no go around options grass strip with deer wandering around on the runway, you better have one of those inefficient statistically accident prone pilots at the controls or the computer will crash your brains out.

IMHO, and I will get flamed about this but oh well: Over the years aviation like the most of the rest of society has made a big push toward being excessively safe and precise to the point that it is becoming too safe to be safe. It's part of the cultural movement toward over micromanaging stuff then defining multiple decimal place computer precision in a limited variable environment as an expectation point for human capability in an infinitely variable environment.

A computer doesn't see or think - it's running blind doing what someone told it to do years before at a desk. A person sees and thinks realtime - and can take a bunch of out of specification variables and do something useful with them.

Most of real world flying is art, not mathematical and computer science.
 
Jaybird180 said:
Do you think that APs have evolved to where they outfly an experienced aviator? Would a human be able to match the precision and finesse of AP flying?
Click to expand...
To some extent that depends on the autopilot (and human for that matter) and what criteria you use to compare their performance. Even the simplest two axis autopilot (when set up and operating properly) can maintain heading and altitude well within ATP testing limits under most conditions as long as the aircraft remains within the required performance envelope (most simple autopilots don't fly well at the ends of the speed envelope). But IME a skilled pilot can fly turbulence smoother than the ones often found in GA airplanes and most any pilot will fare far better than these automation systems when flying a few knots above a stall. Most such units have no airspeed reference and none of the ones approved for light planes have any control over the engine so one clear case where a human will outshine the automation is when there's a reduction of power to the point where a change in the vertical path component is required to maintain flying speed. The latest ones from Avidyne and perhaps Garmin do at least have the ability to recognize airspeed changes and can change pitch in response but most can't even do that and neither of those ones have any effect on the engine output.

Finally there's a whole list of other anomalies such as trim malfunctions, autopilot failures, wake turbulence upsets, miss-programming, etc. which can result in an otherwise properly performing autopilot doing something unsafe (or at least unpleasant) that no sane pilot would do because (most) pilots are capable of reasoning and autopilots aren't.

At the other end of the scale the "autopilots" in something like an Airbus A330 have triple redundancy (ability to detect internal problems and correct them for any single failure), total aircraft envelope protection and some ability to avoid obviously stupid actions but it also depends heavily on proper input from it's sensors, something that Air France found out the hard way. But even in that case the aircraft's flight computer would probably have done a better job handling the emergency had the pilot operating the stick just let go.
 
I think that in the near future there will be general aviation autopilots that 'refuse to crash' into mountains and other vertical obstructions.
 
An auto pilot can be made to fly better than a human.

But a human can be trained to be a better pilot than an auto pilot.

Sort of like the post solo student. Yes he can fly an airplane, but that doesn't make him a "pilot"
 
Jaybird180 said:
Do you think that APs have evolved to where they outfly an experienced aviator? Would a human be able to match the precision and finesse of AP flying?
Click to expand...
I've yet to see an autopilot that'll greet the passengers, load them up, start the engines, communicate with atc, taxi to the runway, depart, fly the flight, land, taxi, shutdown, and deal with the thousands of combinations of emergencies that could occur. So no, I don't think they can out-fly an experienced aviator.
 
I was feeling all bad about my zig-zaggy tracks the other day (no AP in my beast) then I flew a couple hours on the weekend in an a/c with an AP and found myself wondering how folks stay awake while using george! I think I need the stimulus of working the yoke in a piston plane to avoid somnolence.
 
I think if you take a look at the results of DARPA's Grand Challenge competition between autonomous road vehicles, it seems that in another 5 years (and possibly already) an AI autopilot system should be able to exceed a human pilot for an entire flight from start to stop.

Here's the Wikipedia entry on the Grand Challenge:
http://en.wikipedia.org/wiki/DARPA_Grand_Challenge
 
Not that I am in any way qualified to answer this question from an aviation perspective, but: an autopilot will outperform even the most experienced pilot every time under NORMAL conditions.

A pilot - even an inexperienced one - will be better at dealing with the plane under ABNORMAL circumstances, in terms of ending with a favorable result (walking away from the crash).

Since normal conditions exist for the vast majority of our time spent flying, the AP will tend to, over time, give us a false sense of security.

Therefore, while there is an AP in a couple of the planes I fly, the only thing I ever use it for (and usually not even that) is maintaining heading. I am simply not comfortable with the idea of not forcing my piloting skills to be used on a regular basis. This may change with time, I don't know.
 
The new Cessna's with the G1000 and GFC700 digital autopilot are intended to be flown on autopilot from the moment you hit the minimum engagement altitude after takeoff to the moment you hit the minimum engagement altitude on approach.

That's the Cessna/Garmin philisophy reflected in their training materials. And for "mission" flights where the idea is to get from point A to point B, as safely as possible, I agree with this philosophy, particularly when flying IFR.

For flight training, of course, the role of automation is different. It can also be different for other types of flying, such as VFR flightseeing or just the fun of being one with the airplane as you dance through the sky together.

I've flown in airplanes where the autopilot was so badly maintained that it was hazardous to turn it on.

The advice I give to everyone about automation is that if you don't know how the automation works, and why it works that way, so that you can anticipate what it will do when you engage it or change its mode of operation, you'll be better off leaving it alone until you do reach that level of proficiency. I've a friend who is an instructor on "bigger" airplanes used by airlines and business, and he told me the most common thing he heard when people were going through training on the automation was somebody saying "what's it doing now?". That's a perfectly appropriate question when you're sitting in class in a Sim or trainer. It's NOT what you want to hear or say when you're in the clouds with lives on the line.
 
I think Lance hit the nail on the head. There are some autopilots that I can outfly pretty well. In turbulence, I typically turn off the autopilot for my comfort (and that of my passengers). Yes, I can maintain my altitude +/- 20 ft and my heading +/- 5 degrees in significant that makes my passengers cry. But if I allow a few deviations for the sake of comfort, they're less likely to need the relief bags in the seat back in front of them. Or in the case of the dogs, I'm less likely to need to clean out their crates. :)

An autopilot will never do more than a human can tell it to. I think they're excellent tools and I'm happy to have them in the planes I fly (especially over the distances I fly), but there are a lot of days when I just feel like hand flying.

Also, it is important for all pilots to have good stick-and-rudder skills hand flying. I have noticed most Cirrus pilots lack these skills, as they are taught to use the autopilot from 200 AGL back until 200 AGL. When the autopilot fails or something else significant goes wrong, the autopilot won't save you. Overdependency on automation is a problem.
 
I'm pretty sure that I didn't tell my 55X to roll the plane into an 90 degree bank when engaged.
 
TMetzinger said:
I've flown in airplanes where the autopilot was so badly maintained that it was hazardous to turn it on.
Click to expand...

Ha, one of the autopilots in our club planes would crash the plane in about a minute if you just turned it on and let go. It just banks hard left and stays there. Now that I think about it i'll complain because that could be dangerous. It should be fixed or labeled inop/disconnected. Fortunately the first time I found this out I was at altitude and switched it off once we rolled to about 30 degrees bank.
 
Broke autopilots and wing levelers aside, in my experience any AP with Y/D and GPSS will fly smoother and more precise than most pilots. The system in the plane I fly will climb to or descend to a pre selected altitude, level off and hold with in 10 feet of indicated altitude. It will intercept a course with no overshoot and have the right wind correction when it rolls wings level. It will follow the G/S on the money. It can fly to the published hold and enter the hold correctly. It will, if commanded, hold a verticle ascent or descent in vertical speed or IAS. However it will hold that vertical speed to a stall if you do not monitor speed and power. I can not do all of this perfect everytime. In cruise for three hours it is nice. In a busy IFR enviroment I think it is esential. Depends on the mission. The A/P never gets overloaded and will do the same thing over and over with precision. Few pilots can claim that, for sure I can't. JMO
 
Not a single comment asking who was looking outside the window...?

Hopefully while one person is fiddling the other person is on outside scan, and it's a formal hand-off?

I'm still convinced the Cirrus driver that slammed into the Pawnee towing a glider over Boulder, CO was heads-down playing with the gadgets, VFR.

Just a personal opinion... gadgets are great but they do pull your eyeballs inside way too much. Me included. Even in my venerable ol' /A airplane.
 
When the turbulance gets real bad, I hand fly. The autopilot cannot anticipate things, it can only react. Sometimes reactions happen to late.
 
TMetzinger said:
The new Cessna's with the G1000 and GFC700 digital autopilot are intended to be flown on autopilot from the moment you hit the minimum engagement altitude after takeoff to the moment you hit the minimum engagement altitude on approach.

That's the Cessna/Garmin philisophy reflected in their training materials. And for "mission" flights where the idea is to get from point A to point B, as safely as possible, I agree with this philosophy, particularly when flying IFR.

For flight training, of course, the role of automation is different. It can also be different for other types of flying, such as VFR flightseeing or just the fun of being one with the airplane as you dance through the sky together.

I've flown in airplanes where the autopilot was so badly maintained that it was hazardous to turn it on.

The advice I give to everyone about automation is that if you don't know how the automation works, and why it works that way, so that you can anticipate what it will do when you engage it or change its mode of operation, you'll be better off leaving it alone until you do reach that level of proficiency. I've a friend who is an instructor on "bigger" airplanes used by airlines and business, and he told me the most common thing he heard when people were going through training on the automation was somebody saying "what's it doing now?". That's a perfectly appropriate question when you're sitting in class in a Sim or trainer. It's NOT what you want to hear or say when you're in the clouds with lives on the line.
Click to expand...

My initial thought was in agreement, but after the video my utilization of FMC will change.

I definately understood what he meant by reducing automation to reduce taskload.
 
If you can anticipate turbulance, you are very good. I think a pilot will have to react to what the turbulance is doing. You will not be able to react as fast as the AP.
 
ronnieh said:
If you can anticipate turbulance, you are very good. I think a pilot will have to react to what the turbulance is doing. You will not be able to react as fast as the AP.
Click to expand...
The AP in most of our light ga aircraft will try too hard in turbulence and instead of riding out minor changes it fights quickly to maintain the desired profile and as a result isn't very smooth. Some of them (like the Stechs) have a tendency to try and hunt around a bit as well.
 
Jaybird180 said:
My initial thought was in agreement, but after the video my utilization of FMC will change.

I definately understood what he meant by reducing automation to reduce taskload.
Click to expand...

Unless you're flying a G1000/Avidyne or bigger airplane, you're not flying an FMC.

Most GA airplanes only offer the "2nd" level of automation, not the third.

I don't see any discrepancy between the Garmin/Cessna philosophy of having the autopilot fly the airplane and the AA philosophy of choosing the appropriate level of automation.

The problem the AA video addresses is what happens when you get changes (such as ATC starting to give you vectors). Don't try and reprogram the FMS - use the autopilot controls to set the new heading or altitude. You'll notice that AA doesn't say disconnect the autopilot except in limited cases, like a runway change.
 
ronnieh said:
If you can anticipate turbulance, you are very good. I think a pilot will have to react to what the turbulance is doing. You will not be able to react as fast as the AP.
Click to expand...

As Jesse said, the AP will try too hard to maintain a particular heading/altitude/course, and produce a very choppy result. It can do this like an amateur student pilot, with hard inputs that make the passengers unhappy.

An experienced pilot can improve things significantly for the passengers, and will know when it's appropriate to make major inputs vs. when it's appropriate to go with the flow. An AP doesn't know the difference.
 
Ted DuPuis said:
As Jesse said, the AP will try too hard to maintain a particular heading/altitude/course, and produce a very choppy result. It can do this like an amateur student pilot, with hard inputs that make the passengers unhappy.

An experienced pilot can improve things significantly for the passengers, and will know when it's appropriate to make major inputs vs. when it's appropriate to go with the flow. An AP doesn't know the difference.
Click to expand...

Attitude-based units do better at this than the rate-based, but you're still correct, in the bumps, it's often better to hand-fly. In turbulence, many autopilots will hit a limit and disconnect anyway.
 
ronnieh said:
If you can anticipate turbulance, you are very good. I think a pilot will have to react to what the turbulance is doing. You will not be able to react as fast as the AP.
Click to expand...

Does an autopilot take into account its last 100 corrections and use those to predict the needed rate and degree of deflection needed for turbulance hits?

Humans do.
 
Last edited:
There are other times when it is functioning (as designed) the auto pilot can be fun. I departed out of Dulles one day (with a stiff wind so I didn't notice it during the takeoff roll) and the autopilot started doing S-turns across the course. What the *#@) is wrong with this thing. I punched the red button and immediately the plane tried to roll right. One tip tank was pretty darn near close to empty and the other full, OOPS.
 
TMetzinger said:
Attitude-based units do better at this than the rate-based, but you're still correct, in the bumps, it's often better to hand-fly. In turbulence, many autopilots will hit a limit and disconnect anyway.
Click to expand...

You probably fly newer airplanes with nicer autopilots than what I'm used to. Most of my AP experience is with old Century IIIs, but I haven't exactly been impressed with the new S-Tecs I've seen, either. A lot of that may have to do with the plane itself, though.
 
In realy bad turbulance maybe so. Most of my experience is with the KFC 200 without the Y/D. In alt hold it does pretty good. With the Y/D it does very good. The AP I now fly (Collins 65) does very well and has a "soft ride" mode for turbulance , I have never used it. I have never seen anyone that can anticipate the yaw better than a good Y/D.
N2O junkie, no.
The OP asked if a human can fly smoother and more precise (my words) than a correctly functioning AP. With in the scope of the AP abilities I think normally no. Attitude AP does work better than rate based, IMO. Auto pilot use depends on the mission and circustances. For my kind of missions it is almost esential. A 50 mile hop in a 110 knot plane in VFR not so much. Everything in between depends.
 
The funniest thing about our autopilot is watching it do an ILS in turbulence. The yoke is all over the place but the result is fairly smooth. It's not that great at intercepting and sometimes makes a couple s-turns but once it's locked on it will fly the course better than a human, even in turbulence. But the first time I saw the yoke jumping around I thought, holy cats it's working hard.
 
Richard said:
http://www.youtube.com/watch?v=h3kREPMzMLk

Discuss.
Click to expand...



By today's standards of technology, that video (1997) is clearly outdated. It follows on the heels of the 1995 American Airlines 757 crash in Columbia.

With today's "big picture" moving map presentations on panel mounted GPS systems as well as many handhelds...........pilots will know exactly where they are in relation to airports, navaids, and especially rising mountain terrain. I give very little validity, to informational videos such as this, that were created so long ago. It has a lot less relevance today.

L.Adamson
 
Used correctly, an autopilot makes a good pilot better. But it also can make a bad pilot worse. I hand fly my airliner quite a bit, as I don't want it to be an emergency every time the autopilot gets disconnected. Doesn't matter how good the automation if the pilot can't land in a stiff crosswind. Flying any airplane well takes practice, so I practice.
 
Everskyward said:
The funniest thing about our autopilot is watching it do an ILS in turbulence. The yoke is all over the place but the result is fairly smooth. It's not that great at intercepting and sometimes makes a couple s-turns but once it's locked on it will fly the course better than a human, even in turbulence. But the first time I saw the yoke jumping around I thought, holy cats it's working hard.
Click to expand...

Yep. Making lots of corrections per second means that even though some of them may be incorrect the end result is very smooth.
I flew a DA40 XLS last night with the Garmin Autopilot (digital, attitude based). It's so much better than the KAP 140 (Digital, rate based) that the DA40s initially came with it's not funny. I used the flight director the whole time and let the autopilot fly if I needed my hands for something else, but mostly I hand flew. I also did a few legs raw-data without the FD for that practice since it had been a while since I was in IMC.

When the time came for the ILS, I used the FD and hand flew, and worked really hard, probably with similar results to what you describe. Lots of constant small corrections but a very smooth approach. If I'd been flying raw data I'd have made fewer, larger corrections.

And it's STILL cool to see the runway lights appear out of the murk exactly where they should be.
 
You must log in or register to reply here.
Back
Top