Auto Tune Mode

[Leonard Hall]

Hi all,

Randy and I have finished our first attempt at auto tuning. It has given good results so far but it is time start getting feedback from others on their experiences and on the approach we have taken.

The method

We excite the appropriate axis with a step command and then examine the response. For the rate P and D terms we use a step rate command and for the Stab P we use a position step command.

We have 4 steps.

1. We increase rate D until the bounce back or ripple increases past 10% of the commanded rate for 4 tests. At the same time we keep the maximum between 80% and 100% of the commanded rate to ensure that the P output isn't contributing to the bounce back.

2. We decrease rate D until the bounce back is less than 10% for for 4 tests.

3. We increase rate P until the peak increases over 100% for 4 tests.

4. We increase Stab P until the peak increases past 110% of the commanded angle or the bounce back increases past 10% of the commanded angle. We then assign the minimum of the roll and pitch axis time 0.75 as the tuned value.

To test auto tune yourself you can find it in trunk. The way to try it out is to:

1. open the latest trunk ArduCopter code in the Arduino IDE and add this line to APM_Config.h:

     #define AUTOTUNE     ENABLED

     #define MOUNT            DISABLED

2. upload the code to your APM

3. set the CH7_OPT or CH8_OPT to 17 to allow turning on/off the auto tuning with ch7 or ch8

4. ensure the ch7 or ch8 switch is in the off position

5. take off and put the copter into AltHold mode at a comfortable altitude

6. put the ch7/ch8 switch int the upper most position to engage auto tuning:

     - you will see it twitch about 20 degrees left and right for a few minutes, then it will repeat forward and back

     - use the roll and pitch stick at any time to reposition the copter if it drifts away (it will use the original PIDs during repositioning and between tests)

     - move the ch7/ch8 switch into the off position at any time to abandon the autotuning and return to the origin PIDs

7. when the tune completes the copter will be left with the newly tuned PIDs active, fly around to see how it responds

8. If you are happy with the new PIDs:

     Leave the ch7/ch8 switch in the ON position, Land disarm to save the PIDs permanently.

     If you DO NOT like the new PIDS switch ch7/ch8 off to return to normal PIDs.  the gains will not be saved when you disarm

Note: if you have a 3-position ch7/ch8 switch you can interrupt the tune at any time and test out the PIDs so far by putting the switch in the middle position

Thanks all and I look forward to your feedback.

Leonard

[Leonard Hall]

Hi,

I should also say that when you are doing this for the first time you should wait for a calm day and go to a wide open area.

The tuning only tends to move the copter a few meters each way but because you should be trying to not touch the sticks as much as possible, the copter will drift with the wind. While you can do the tune during mildly windy conditions you will need to get very good at quickly moving the copter back to it's starting location.

When starting from a default tune you should expect to need around 6 to 8 minutes if you don't need to take over very often. If you do need to take over quiet a bit then it can take 10 or 12 minutes. This is getting pretty close to a full battery on the standard 3dr quad with 4500mAh batteries.

Leonard

[Robert Lefebvre]

Cool, I'll try to test it for you.  Couple questions:

How much room do we need to do this?

Can we nudge the copter around to keep it in position?

--
You received this message because you are subscribed to the Google Groups "drones-discuss" group.
To unsubscribe from this group and stop receiving emails from it, send an email to drones-discus...@googlegroups.com.
For more options, visit https://groups.google.com/groups/opt_out.

[Leonard Hall]

Hi Robert,

At any time you can take over using the sticks and it effectively switches to stabilize mode using the default PID settings and you can fly normally. Once you let go of the sticks it will continue the test again.

Thanks!

[Robert Lefebvre]

Oh I see, nice.

So it sounds like I just want to go to a big area, and just let it go?

Is it obvious when the procedure completes?

--

[Emile Castelnuovo]

Leonard, that is a great feature and can't wait to try it.

Only one question: why is that there is no I tuning.

What are the reasons about it?

I remember you talked something about I term being the same of P, but I can't understand the reaon why this should or cannot be included in the tuning process.

Cheers,

Emile

2013/10/4 Robert Lefebvre <robert....@gmail.com>

[Leonard Hall]

Robert, you will know when it finishes because it goes through roll, then pitch, then stops. In the end we should make it beep or flash lights or something but for now it just doesn't do anything else.

Emile, The I term is set to react to the additional toque of propeller flapping during fast forward flight. By setting it equal to P we get a reasonably fast response to the additional torque but without increasing the overshoot. So the I term isn't chosen based on the performance of the copter as it rolls and pitches like P and D. It is chosen based on how the copter reacts as it speeds up and slows down. My theory is that by setting it to be equal to P it will result in a reaction time of 1 to 2 seconds. (this could be made faster at the expense of bigger overshoot on roll and pitch rate).

[Marco Robustini]

Fantastic update Leonard, you are the man! ;-)
It would be great if the repositioning of the quad happen automatically during auto-tuning, ie, if the quad during procedures out more than a certain number of feet from the point where it engages the procedure the code switches to Loiter, replace the quad, back to Alt-Hold and the procedure continues from where has interrupted... i'it feasible? :-D
I know that APM code is already too heavy, but on the other board with STM32 (Pixhawk and VR Brain) the code space and processing power are not lacking!

- Marco -

[Dirk Milz]

@Marco:

Don´t forget to switch on the Cam when testing it. ;-)

My APM is updated but due to wind I couldnt test it today.

If you tested it, please let us know (and see) the space you need for testing.

And of course your PIDs before and after.

Thanx

Dirk

Can´t wait to test it on my custom homemade frame....

[Robert Lefebvre]

I'm really curious to see if it works well on my Nerf Quad.  It's a bit of an odd-ball.  A flexy 450 frame with 12" props on it (way too big!).  I haven't been able to get a great tune out of it, and it's a bit unstable.  Also, the APM damping is fairly soft so that might give funny results. So I can't wait to see how autotune does.

Does the autotune depend on a reasonable starting tune?  What happens if the tune is really bad to start with?

[Leonard Hall]

Marco:

Yes using loiter is something I would like to do. I was thinking of using loiter if auto tune was engaged from loiter. Basically between tests it would ask for zero velocity from the loiter velocity controllers but then if it gets more than 5 meters from the auto tune engagement point it would to a return to... Engagement point.

There are other things I also should do like freeze the alt hold throttle output during the tests.

This will all get much easier when we rewrite the code to clean up the attitude stuff.

[Rick Payne]

I gave it a go on my test quad, which is quite overpowered. It took about 6 minutes to do with a couple of repositions and the result is as flyable as my manually tuned PIDS. FWIW, here's what it got - 1st column are mine, second column the autotune results:

RATE_PIT_P             0.0600       0.0450
RATE_PIT_I             0.1000       0.0450
RATE_PIT_D             0.0040       0.0035

RATE_RLL_P             0.0600       0.0500
RATE_RLL_I             0.1000       0.0500
RATE_RLL_D             0.0040       0.0020

So it definitely gets the thumbs up from me - great work guys ;)

Cheers,

Rick

[Jason Short]

Does the intern default to 1x the p term?  Would be interesting to see the I term tuned after auto tuning to see what effect you get. My guess is a higher I will increase the locked in feel. 

Jason

Sent from my iPhone

[Robert Lefebvre]

Not having a lot of luck so far.  I tested it out on the Nerf Quad, which is a bit peculiar as the props are too big for the frame. I've got it flying OK, but not great.  Sort of as I expect for any large prop slow motor machine.

Anyway, I believe the auto tune completed.  I found it left me with very different settings between pitch and roll.  Pitch was OK, but roll was somewhat unstable.  Very sloppy.  I wanted to see what the numbers are, so I tried to land with the Ch7 switch up so I could do a disarm and refresh to see the PIDs.  But soon as I touched the ground, it bounced up and flipped over and crashed upside down.

I disarmed, still with the Ch7 stick up, and did a refresh, but it's still showing my old PIDs?

[Robert Lefebvre]

Ok, if I put Ch7 high, then do a refresh, I think I get the new numbers.

My Numbers:

P: 0.07

I: 0.07

D: 0.006

Auto Tune

Roll:

P: 0.16

I: 0.0016

D: 0.006

Pitch:

P: 0.07

I: 0.07

D: 0.006

So, the pitch numbers are surprisingly close to mine.  I'm not sure if it had an effect on the process, but I'm doing it in my back yard, I have about 50' square, I had to take over control fairly often to reposition it.

[Robert Lefebvre]

After doing a reboot, the parameters are now:

P:0.045

I: 0.045

D: 0.005

I think I've broken the procedure! :D

[Leonard Hall]

Hi All,

Great to see the testing begin.

Could you please send me the logs of your tuning flight. The logs let me see the measurements of each test as well as the results so I can very easily see what is happening.

Rick: Well done on your manual tune. What did Stab P come out as.

Robert: Logs will tell me what is happening here. The only thing I have found makes the tune fail (so far) is a combination of soft mounting for the apm letting the apm wobble as the tests are done.

If you don't like the tune you get I would suggest just looking at the logs rather than continuing to fly with it.

Thanks all,

Leonard

[Leonard Hall]

Oh, and the logging for tuning doesn't need to be turned on. :)

[Robert Lefebvre]

Oh yes, the Nerf quad has fairly soft mountings.  That could be it.

[Robert Lefebvre]

[Julien Dubois]

Great job!

Loaded it but couldn't finish the test because of strong wind.
You'll get my feedback asap.

From now, I confirm we can cancel the autotune or retrieve control by moving stick very well, so there is quite no risk testing it!!


Btw, is there a section or document explaining all the different variables of the project and how the copter is modelled and the control loops are made?
I'd like to make some change in loter and alt hold behaviour and it's quite hard to reverse engineer the C code.

Julien

[Leonard Hall]

Thanks Julien, 

As for the question you ask, Alt hold is pretty straight forward and I have done some diagrams on that one but Loiter is VERY complicated and primarily just mathematical. But in it's most basic form, Loiter is a three dimensional version of Alt Hold. If you start a new discussion and go into a bit more detail on what you want to do I can try to help more.

Leonard.

[Robert Lefebvre]

Leonard, further to our discussion last night, do you think that Auto Tune could be used to tune Rate Feed-forward?  I believe FF would benefit multicopters, but it's been resisted because nobody wants to have to tune another parameter.  Maybe we could bring in FF but not require people to have to tune it.  Basically, you would start the auto-tuning process with FF at zero, and run through the PIDs as before.  But when done, you would then look at how much P+I is being required to achieve the rate, and then start working on an FF based on that number?

--

[Jonathan Challinger]

I've tried FF on multicopter. It isn't necessary or reasonable

[Robert Lefebvre]

Really?  More detail please?

[Leonard Hall]

While I am sure I can autotune a rate feed forward, A stab feed forward we can calculate the coefficient exactly.

However, the roll and pitch response is so fast I don't think we could tell that it is there.

I think we would get a lot more mileage doing some input conditioning to limit the maximum rate the angle changes. This would let us tune up the kids as sharp as we want but still have a docile feel.

[Robert Lefebvre]

Yes, at the least I would like to do that.  As it is now, I was flying my H8 with Stab P at 3.5.  This was to achieve a nice feel, but I'm sure it did not resist outside disturbances as well as it could.  

[Leonard Hall]

Randy and I have made a small change to the Auto Tune to allow some copters to tune that require a very low RateP term. We have reduced the minimum RateP to let it go as low as 0.02 during RateD tuning. this is done to ensure there is no overshoot while RateD is being tuned.

We have also made the Auto Tune complety stop if the minimum value of RateP is reached.

Thanks,

Leonard

[vanta...@gmail.com]

I downloaded the ZIP file to give rc3 and autotune a try.  I'm unable to compile the file either before or after adding the Autotune commands.  I'll try attaching a screen shot of the error I'm getting. Perhaps someone can help me get past this.  I tried with Arduino 1.0.1 and 1.0.4 with the same results.

[Randy Mackay]

Vantassic,

     You'll need to use the modified arduino.  There's instructions on this wiki page: http://dev.ardupilot.com/wiki/building-the-code/

On Thursday, October 10, 2013 12:43 PM, "vanta...@gmail.com" <vanta...@gmail.com> wrote:

I downloaded the ZIP file to give rc3 and autotune a try.  I'm unable to compile the file either before or after adding the Autotune commands.  I'll try attaching a screen shot of the error I'm getting. Perhaps someone can help me get past this.  I tried with Arduino 1.0.1 and 1.0.4 with the same results.

[Rick Payne]

Sorry for the delay, life keeps getting in the way of the hobbies ;)

Stab P values were the default I think (4.5 anyway) and they were tuned to 5.8125. Log is attached...

[Andreas Baumgartner]

Hi all,

 

First of all I have to say, great job!!

 

I tested autotunig about 10 times with hard mounted APM and with soft mounted APM.

All the results have been nearly the same.

Compared to my hand tuned parameters, they are all much more on the lower side.

Flying with autotuned parameter is very flabbily.

I am flying a quadro like the 450 flameweel with 4s and 10x4,5 carbon props, weight is about 1000g.

Parameters are:

                       handtuned     autotuned

RATE_PIT_D: 0,006            0.004
RATE_PIT_I:   0.4                0.05
RATE_PIT_P:  0,12             0.05
RATE_RLL_D: 0,006           0,004
RATE_RLL_I:  0.4               0.05
RATE_RLL_P: 0.12             0.05

 

I think that there is a big difference between hand tuning and auto tuning

and no improvement in flight behavior.

Have you experienced the same?

 

Regards,

Andreas

[Robert Lefebvre]

Ok Leonard, so I was going to put smaller props on this thing, and maybe firmer dampers.  Just cause it's the right thing do to.  But if you want I can test it as-is again?  Is that important info?  See if you've fixed it, or maybe this thing is just too much of a mess.

[Jason Short]

Did you increase Stab_P? That might make up for the lower Rate_P.

Jason

[Leonard Hall]

If you could try again with the same setup it would be great! The results will be interesting!

You received this message because you are subscribed to a topic in the Google Groups "drones-discuss" group.
To unsubscribe from this topic, visit https://groups.google.com/d/topic/drones-discuss/IAxYUZptiZE/unsubscribe.
To unsubscribe from this group and all its topics, send an email to drones-discus...@googlegroups.com.

[Leonard Hall]

Could you please send a log of your tuning flight. The failed tunes are the ones we learn things from.

Thanks!

[Leonard Hall]

Jason the auto tune tunes Stab p as well so if it is sloppy after the tune, the tune didn't work and there isn't shit way to fix it.

You received this message because you are subscribed to a topic in the Google Groups "drones-discuss" group.
To unsubscribe from this topic, visit https://groups.google.com/d/topic/drones-discuss/IAxYUZptiZE/unsubscribe.
To unsubscribe from this group and all its topics, send an email to drones-discus...@googlegroups.com.

[Leonard Hall]

Hi all,

Sorry for the short replies I was using my phone before.

I currently have to ways that the AutoTune can fail. One of them is caused by wobble in the copter that creates false positives in the tune algorithm. I had this problem with Randy and I hope we made it tolerant enough to ensure that this isn't a problem unless it is a seriously bad wobble.

The second we saw with Roberts tune where the minimum Rate_P is simply not low enough for the auto tune to reach point where it isn't overshooting. From what we can tell this is being caused  by using large props on a small frame. But we will see how it goes when he tries again with the new code.

The most important thing to remember is that I need a copy of the logs to see what is happening with auto tune if it doesn't work. The cases where it doesn't work are the most interesting for me and where I must focus my efforts to ensure Auto Tune works for as many people as possible.

Andreas, it looks like you are having the same problem as Robert where Rate_P is reaching it's minimum value and messing up the tune. Are you using a post October 9th version of Master? If you send me a copy of your logs I should be able to quickly see why the tune isn't working.

That brings me to another point. Auto Tune automatically logs each test so I can easily see why it isn't working and what the problem is if I have the logs. So Logs are really important!

Thanks all!!!! So far we have 8 copters that have had successful auto tunes and 2 that have not. While 20% isn't bad, I would like to get that last 20% working if I can.

Chat soon,

Leonard

[Andreas Baumgartner]

Hi,

 

I tested Autotune once again.

May you please explain what exactly I have to do

to save the new parameters permanently.

Im landing with Ch8 option most up and disarm manually.

Are the parameter at this time saved or is it necessary to write them

into eeprom by using mission planner?

Because after only disarming when I switch ch8 back to lowest position, and do a refresh

in MP then all the parameters are changed to very low values, almost not

useable for flying.

My second question is:  is the result after autotuning dependable of the parameters

as they are before activating autotune?

 

Besides I have attached my last autotuning log file...

 

Regards,

 Andreas

 

[Leonard Hall]

Hi Andreas,

It looks like you have the wobble problem that causes autotune to get false positives from the tests. Could you tell us how you have dampened the APM and maybe include some pictures. This problem can be caused by very soft dampening, a loose stack, or heavy wires connected to the APM.

To answer your question if you land and disarm with your 3 position switch set to position 2 or 3 it will save the pids. But you can test fly the new values by setting your 3 position switch set to position 2 or 3 without needing to land and disarm. That way you can switch to position 1 to get your old values back.

Thanks for this Andreas, this is an important data point for me!!!

Leonard

[jnjo...@gmail.com]

Hi Leonard.
Autotune look very promising and I've heard several success-stories in other forums.
Unfortunately it's not going really well for me.
I made two attempts yesterday and ended up with a very sloppy 'copter, with Roll_Rate_P down to 0.05 from the initial 0.190

Daylight disappeared so I was unable to make a comprehensive test with more logging enabled.

One issue I had was that the tuning stopped, apparantly mid-way on several occasions.
I think the reason for that could be a noisy potentiometer in my TX-stick, because when I re-positioned the ´copter and let go of the stick, tuning continued.
My TriCopter is described here -> http://www.rcgroups.com/forums/showthread.php?t=1848266#post24373177

Attaching the two logs and hope they might be useful to you. Please let me know if there is any more specific data I can provide.
I'm not asking for any fault-finding, (although I will happily accept it. ;))

BR,
Jan

[Robert Lefebvre]

I had that happen too.  I repositioned the quad, but the tuning didn't restart.  So I thought it was done.  I moved it again, and then the turning started up again.

--

[Julien Dubois]

Hey,

the wind has finally gone :)

I'm not sure the tuning was 100% finished it last about 2'30. P & I parameters have changed, but not D.

Before autotune



After autotune



The video of the autotune (bad quality, but just to show how it works).
http://youtu.be/ICR6qz5LNnc

Julien

[Leonard Hall]

Thanks Jan!!!!

I will have a look at this tonight and hopefully be able to shed some light on the problem.

I too have noticed that sometimes the autotune stops and I need to give it a little shove with the sticks to get it going again. I suspect that it may be because I am repositioning in mild wind and with the reduced I term it may be holding a slightly larger angle than the AutoTune will allow before continuing.

Chat soon!

Leonard

[James Li]

how about safety issue? the entire auto tune procedure introduces the copter in an unsteady/stable and somewhat erratic state. 

[Robert Lefebvre]

I didn't find it unsafe at all. I was actually pleasantly surprised. Though I can see how a new pilot would find it harder to keep control of, which is sort of unfortunate because they are the ones who need to use the process the most.

On 13 October 2013 19:33, James Li <jy0...@gmail.com> wrote:

how about safety issue? the entire auto tune procedure introduces the copter in an unsteady/stable and somewhat erratic state. 

--

[Andreas Baumgartner]

 

Hi,

 

Next Autotuning Test,

Weather condition are very good.

 

After setting ch7 to most upper position autotuning procedure

starts and only 2 or 3 repositioning are necessary.

After finishing I am landing and manually disarming.

During this Ch7 stays all the time in most upper pos.

 

I refreshed parameters in MP but what a surprise...

none of the PID parameter changed.

 

I was starting with

P 0,1200

I  0,,4000

D 0,0060

 

After autotuning exactly the same values.

 

FW version is 3.1 rc4 (last trunk version downloaded and compiled

on Okt. 13, 9.00 CET)

 

I have done this twice and exaclty same behavior.

Last log file is attached...

 

Regards.

Andreas

[Robert Lefebvre]

Hey Leonard, I tried again today using trunk on my Nerf Quad, see if I could get a better auto-tune, but it did not go well.

First, the process seemed to stop quite often, all on it's own.  I had to keep flicking Ch7 off and then back on to get it to restart.  After a while, it started to get CRAZY soft and the roll axis was just rolling around like crazy.  I decided I'd better land it, so I turned Ch7 off again, and tried to land and found that I was left with the same crazy soft control on the roll axis.  It was a bit sketchy getting it down.  I found that it had left me with P and I of 0.015.  

I'm not sure why I didn't get my default params back when I turned Ch7 off?  That's not good.  I wonder if it could have something to do with the recent changes to the scheduler?

I wish I could get you logs, but I had to turn off CLI to get it to fit in when I compiled (forgot about ctrl-Q in MP) and I can't seem to download the logs.

So, I then changed to the next stiffer dampers for the APM.  It's fairly firm now, I don't believe there would be any APM wobbliness anymore.  I tried autotune again, and had the exact same bad result so I aborted again.  

I then decided to see if I could tune up the Rate PIDs for better control.  I found I couldn't really tune P up anymore. Then I decided to try increasing Stab P from 3.0 to 4.5 (I had had it on 3.0).  This was not good at all.  Very bad.  In Alt Hold, if I pushed it to get it moving, then released the roll command sharply, the quad very often tried to flip right over. I neglected to try tuning up Rate D and realized I should try that as I write this.

This reminded me of the problem somebody else recently reported.  I'm starting to wonder if we're seeing a problem with large/slow prop machines.  They just cannot be asked to change direction that fast.  I put Stab back to 3.0, and flew another battery with no serious issues.

Next, I plan to drop the prop size on this 450 from 12" to 10" and see if it works better.

--

[Leonard Hall]

Hi Andreas,

It looks like you have a fair bit of vibration on your roll and pitch that is causing false positives and therefore bad values during the tune. The Auto Tune found that Stab P was getting too low and aborted the tune and restored your parameters to default. This was done because the Pitch Rate P dropped to 0.025.

Could you send a couple of pictures of the copter and maybe turn on the ATT logging if you try again.

Thanks mate, I really appreciate the time you have taken to test this. I don't have many example of failed tunes so every one counts!

Leonard

[Leonard Hall]

Hi Robert, I am beginning to think you might be right. Shame about the logs though!

[john...@gmail.com]

Yes, there is a limit to how quickly a big heavy octo with 15" can respond compared to a lightweight quad with smaller propeller, so the tuning should take this into account.

But it's the same (inertia) slowness that makes big octos perfect for photo and video work. Such birds are all about making movements as smooth as possible and still be locked in when hovering. This usually means lower P values then normal and more aggressive D's (compared to the low P) to make the big motors respond better.

- JAB

[Robert Lefebvre]

Actually John, this was on an F450.  It's just it had 12" props, normally this frame would use 8-10".  

[Robert Lefebvre]

I'm wondering if maybe the problem with the auto tuning has to do with the fact that the auto tune is done in Alt Hold.  Whatever the root cause of the instability is (large props, etc.) I find that the situation is much much worse in Alt Hold rather than Stab mode.  

I will upload the code again using ctrl-Q so I can leave CLI enabled.  That way I can get the logs for you to see what's failing.  Then try reducing prop size to 10 and see how much difference it makes.

I'm still a little concerned however about the fact that the auto-tune seems to stop on it's own, and that I didn't get my original PID's back when I turned Ch7 off?

[Andreas Baumgartner]

Hi Leonard,

 

Again do an autotunig but none of the PIDs changed...

It seems that autotuning procedure ends correctly, not with an abortion.

May you confirm by checking log file?

Actual logfile is attached.

 

From now on I activate ATTITUDE_FAST as you told.

New logfiles will come soon...

 

Besides, Sonar is not working since 3.1_RC1

although testing sonar in CLI is ok.

 

Regards,

Andreas.

 

[Leonard Hall]

That sounds good Robert,

I think you may need to wait for RC5 (this weekend I think) to download it using the mission planer because it wasn't turned on in that code.

Why don't you disable mount instead?

I will look through the code to see if there is any way that the auto tune values could get saved after an abort tonight. I agree this isn't something we want to happen!

Chat soon,

Leonard

[Leonard Hall]

Hi Andreas,

Yes I can confirm that the AutoTune is aborting near the end of tuning PItch, that is why it looks like it is finishing.

You  can see you are getting significant wobble during the tests and that is why the autotune is failing.

Could you send some pictures of your setup and look for anything that can move the APM on the frame.

Thanks and chat soon,

Leonard

[Glenn McLelland]

Thanks for this feature Leonard. I've had a mixture of good success, and not so much success with auto tune so far.

It worked well (very well) on my symmetrical 700mm diagonal UAP2 quad which is ~2.2kg w/2814-11 motors and APC 12x4.5MR props. Previously I couldn't get a good tune out of it at all, whereas after the autotune it handles like a freaking ninja. PIDs ended up with Rate P 0.19 I 0.19 D 0.02. The Stab P of 9.2 it came up with was far too high for my liking, so I've put it back down to 4.5 and I'm very happy with how it feels now. I'll have to take it out in some strong wind to compare with before, but I have high hopes. In all my previous tunes I didn't have my Rate D up anywhere near high enough.

Where it didn't work so well was on my QAV540g. This frame is asymmetrical, and it has a lot of inertia on the pitch axis as it's very long, and very little inertia on the roll axis. Auto tune didn't complete on this frame despite several tries. It does what feels like a good job of tuning the roll axis, but on the pitch axis it just keeps going higher and higher with the PIDs until it got to the point of a fast continuous twitch in flight, which is where I brought it down to save against a crash. The PIDs it ended up with at that time were: 

Rate Roll P 0.105 I 0.105 D 0.0045, 

Rate Pitch P 0.52 I 0.52 D 0.0455, 

Stab Roll P 7.12

Stab Pitch P 14.5

I've attached a couple of logs from the QAV tuning flights

[Robert Lefebvre]

Oh shoot, I thought it was turned on in trunk now.  I just tried it with ctrl-q, no joy.

Yeah, I could have disabled mount, but I also wanted to test the processor loading.

[Robert Lefebvre]

Well, I couldn't wait anymore, and changed the props down to 10"  It flies WAY better now.  I retuned the PIDS, and ended up having to reduce P from 0.07 to 0.055, and D from 0.006 to 0.0035.  I then cranked Stab P up from 4.5 to 10, and finally settled on 8.  The quad is a completely different machine now.  Much, much more stable, and it's a rocket ship at full throttle.

I still think there is definitely a connection between Alt-Hold usage, and Rate performance.  In AH, Stab P of 10 could cause some nasty overshoot and wobbles.  But in Stab, it was totally fine.  That's why I reduced it to 8.

On 15 October 2013 17:28, Leonard Hall <leonar...@gmail.com> wrote:

[Leonard Hall]

Hi Glenn,

I have seen this once before I think. Could you please tell me the relative placement of your frame and exactly where the CG is. My theory is that the V style frames have coupling between the two axis and this causes the shape of the rate return to be strange.

I need to understand this if I am to have any hope of fixing this problem. If you can help me with the above information I would be very grateful!

Thanks,

Leonard

[Glenn McLelland]

Hi Leonard, Sorry accidentally replied directly, but I think the issue with the QAV is that it it's not a V layout, it just has the weight concentrated at the front and back so it takes a lot more motor input to make a pitch movement than a roll movement. 

[Leonard Hall]

Ok, maybe we can take this off line and if you are willing see if I can make some code changes to handle this weight distribution.

Chat soon,

Leonard

[Robert Lefebvre]

Glenn, do you know if the CoG on that frame is roughly close to the geometric center of the motors?  Or is is more forward or backwards?

[Dirk Milz]

@Leonard and Glenn:

Please discuss this here and not privat. Many of us using these Frames where Cam is in the Front and Battery in the Back. Me too.

Have same problem like Glenn and I think most others flying FPV with these Frames. Advatage is that Props are outside the Picture.

So, please discuss it here (or open another Thread).

If you need more Testers for that, my Frame is ready to fly. Actually flashed with -RC4

Cheers

Dirk

[Leonard Hall]

Dirk,

I am taking this off line because I have a couple of things I would like to try with this particular tuning problem. I also want to talk with Glenn over skype to get more effective communication going about the problem.

Hopefully my simple fixes (a cheat really) will work on his frame and then we can report back here what we did and what worked and what didn't work.

Have you done an auto tune. If you have a similar frame to Glenn, I would be very interested to see if you get similar results. If you do it would be great to have another tester to see if my changes effectively solve the problem.

Thanks and chat soon,

Leonard

[Rick Payne]

Leonard - my successful autotune was on a very similar-style frame. I sent the logs for that before..

Cheers,

Rick

[airm...@gmail.com]

Hi Leonard,

Following our conversation here http://diydrones.com/xn/detail/705844:Comment:1437863, I made a new test tonight, and it was successfull.

The first test was done with rc3, the second one with rc4.

The main differences between the 2 tests are :

- Change version rc3 vs rc4,

- In rc3, I made a small modification in auto_tune.pde, I took this one : https://github.com/diydrones/ardupilot/blob/7e657ce856722edf9f7189db6606806958525377/ArduCopter/auto_tune.pde

- For the first test, there was a little more wind than second one, so I had to adjust the copter position more oftenly (maybe a bug here)

I confirm that when you want to pause Auto Tuning (CH7 or CH8 switch in middle position), tuning restart from the beginning.

I can't provide you any logs, because I was Auto tuning with no log activate (I know no need to do that to have Auto Tune logs), but after successfull flight I enabled logs to see the difference and then the Auto Tune log were automatically erased ...

[Leonard Hall]

Hi Rick,

Yep, I found the file but you zipped it up using gzip or something strange. Could you resend the file as just a uncompressed log file. This is the file name: quad-2013-10-06-1.log.gz.

Thanks,

Leonard

[Leonard Hall]

Hi mate,

What values did you end up with?

[Robert Lefebvre]

I gave this a shot again, compiled from trunk to enable Autotuning, but I had to turn off Mount, Optflow, Sonar and CopterLeds to get it to fit.

This time with 10" props, much crisper response, but again it seems to just stop for no reason?  Click Ch7 off, then back on, and it starts up again.  Then it started raining so I couldn't let it finish.

[Leonard Hall]

Robert,

Do you have logs?

[Robert Lefebvre]

Sure, here's the one from tonight.  Can you see when I turn Autotune on and off?  I had to reposition it a couple times, but it definitely stopped on it own once or twice.

[Robert Lefebvre]

Yeah, you can see it just stopped at line 1375, and then I waited for a while, but it was stopped.  So I took the opportunity to reposition it.  Then I waited for a bit, it didn't restart, so I think I flicked Ch7 low then high again, so it restarted on line 1640.  I think it stopped on it's own again about 2340, and that's about the time the rain started so I had to land.

[Leonard Hall]

Yeh, it did. This is still exactly the same problem as before. I am thinking this is being caused by the dampening.

[airm...@gmail.com]

Hi Leonard,

I end up with the following PID :

          SP     RP      RI     RD

Roll  : 6.515  0.064  0.064  0.003

Pitch: 6.515  0.104  0.104  0.008

Before auto tuning, my copter was very stiff, sometimes nervous, now it's more FPV like, smoother. Before when flying 30-40 km/h, in FPV I could see the horizon line moving up and down, now it's rock solide.

Thanks for coding this, it's a really nice feature.

Do you plan doing this for other tunable parameters (Throttle Rate / Altitude Hold / Loiter ) ?

[airm...@gmail.com]

In the comment of auto_tune.pde there's :

#define AUTO_TUNE_PILOT_OVERRIDE_TIMEOUT_MS  500    // restart tuning if pilot has left sticks in middle for 2 seconds

I think it should be AUTO_TUNE_PILOT_OVERRIDE_TIMEOUT_MS 2000
To have 2s, because actually it's only half a second :)

[Leonard Hall]

Yeh, you are correct but we actually need to change the comment to say half a second :)

[Leonard Hall]

And thanks for the feedback and I am glad your tune went so well !!!

[Leonard Hall]

HI airmamaf,

I feel a bit silly. I have been trying to find your post on the 3.0.1 thread but had lost you then I found all you posts but didn't put 2 and 2 together.

I am glad you had a good result but I am not sure why though. I think the only change to autotune is an exit if some of the parameters get too low.

Do you have a log of your successful autotune? And have you changed your frame in any way between the two autotunes. For example have you secured your battery differently?

On an unrelated note I have finally found where people are getting this habit of drawing a line between opposite motors to calculate the center of thrust (and then generally thinking this should be the center of gravity). At least they correctly define the CG even if they got the CoT wrong. So the CG is in the correct place.

This is fantastic that you have been successful tune, now we just need to work out what has changed.

Sorry again for getting so mixed up with who you were and being unable to find your 3.0.1 post!!!!

Chat soon,

Leonard

[airm...@gmail.com]

Hi Leonard,

There was no frame change between successfull and no successfull test, only a change of release. Everything else was the same, except less windy day for the successull test (only one repositionning) than the no successfull one.

I did 2 successfull tests, the first one lead to PID maybe to low, because after a hard movement, my copter was about to reverse, but the second one change this behavior.
Sorry, but I don't have any log, they were erased after I decide to turn logging on after Auto tuning.

[Leonard Hall]

Airmamaf,

Maybe it was simply the additional wind.

Do you know what release you were using the first time?

I really appreciate you taking the time to give me this feedback!!

By the way, you have a very impressive frame there!!!

[Robert Lefebvre]

Leonard, are you saying that the skewed-X drawn on the trapezoidal frames does locate the CG properly? (ie, towards the rear)

--

[Robert Lefebvre]

Ah, I found the context.  I see what you're saying.  I actually tried to tell people that the CoT on the TBS instructions is wrong, but that didn't go over well.  As you say, TBS have at least drawn the CG in the correct place, so it doesn't really matter that they have the CoT wrong, though it unfortunately spreads misinformation.

[airm...@gmail.com]

The first test was done with rc3, and I made a small modification in auto_tune.pde, I took this one : https://github.com/diydrones/ardupilot/blob/7e657ce856722edf9f7189db6606806958525377/ArduCopter/auto_tune.pde

Thanks for the frame's compliment :)

David

[airm...@gmail.com]

I made a new test tonight, and it's also working fine. Attached you'll the logs

[xerr...@gmail.com]

Hi,

I tried this today on a quad. It seemed to work very well and I felt it was very safe to do. I did get some strange results,

First attempt seemed to go well, I landed with ch8 still on and disarmed. refreashed my params and I saw no changes. so I tried again

second attempt was the same, went well, landed, disarmed then switched off ch8 (I never switched it off on the first attempt)  but still no changes in my PIDs.  I disconnected the battery and reconnected to the MP and there were changed PIDs but they were far from my PIDs so I thought I would try again

third attempt went the same as the first two, all seemed to go well but no changes in my PIDs till I reboot (disconnected and reconnected my battery) the changes were quite a bit different than my manual PIDs also I noticed my I terms changed too  I did not think that should happen. I will attach a pic of my original pids (in the "new" column)  It surprised me how well the new autotuned pids are flyable and not too bad but a bit sluggish compared to what I am used to. still I had never tuned my D or I terms.




If you want the logs let me know,

thanks for any input on the reesults

John

[Leonard Hall]

Hi John,

It would be great if you could attach logs. That way I can see if the autotune was working correctly on each flight.

I am interested that you couldn't see the pid values had changed until after you power cycled. Could you describe exactly how you were checking this?

Thanks for your help!

[Robert Lefebvre]

I've had the exact same issue with not seeing the PIDs until after a refresh.

I landed, left Ch7 high, disarmed, then dropped Ch7.  Then did a "refresh params. The old params were still showing.  After rebooting, I get the new params.

[xerr...@gmail.com]

Hi,

I did the same as robert only I was using ch8.  Landed and disarmed in althold ch 8 still on (high) then went to MP still connected refreshed and old params still showing.  On the second try I switched ch8 off (low) after disarming but still had the same thing.  I had even tried a preflight reboot on the first try. I dont know if it actually reboot but the numbers did not change till I removed the battery and reconnected.

here are the logs I did not have logs enabled on the first try, I had logbitmask to "0" but logs were enabled in the compile.  I do have optiflow and mount disabled in the compile.

John

[Brandon Basso]

Has anyone experienced the PIDs NOT being saved after landing and disarming with CH7/CH8 on?  I went through the full tuning procedure with an X8, had to make some position corrections along the way, but the procedure more or less went fine.  I tried twice.  Attached are the before/after param files.

--
You received this message because you are subscribed to the Google Groups "drones-discuss" group.
To unsubscribe from this group and stop receiving emails from it, send an email to drones-discus...@googlegroups.com.
For more options, visit https://groups.google.com/groups/opt_out.

--

Brandon Basso, PhD :: Senior Research and Development Engineer :: 3D Robotics :: Berkeley, CA

[Brandon Basso]

Video:

http://www.youtube.com/watch?v=3mMvSmSyCaY

[Leonard Hall]

Hi Brandon,

Could you provide the flash logs for each of those tunes?

Thanks mate!!

Leonard

[xerr...@gmail.com]

Leonard,

I tested autotune today on a different copter, its an octa-quad. all went perfect, end of test I disarmed went to MP and refreshed params and the new numbers were saved.  it was on a quad that it was not saving yesterday.  I have one other apm25 based quad I will try tomorrow if I can. I will let you know how it goes.
weird, I did everything the same as I did yesterday but it worked perfect.

[Leonard Hall]

Ok, I think I know why some people aren't getting having their gains saved. The auto tune is exiting because the Rate P has reached the lower limit and the AutoTune has aborted.

The next release will have the minimum reduced back to what it was initially (from 0.05 to 0.02). I just don't want to cause a crash by allowing a value that is too low.

I am testing the AutoTune that will be in the next release today to make sure it works as we expect.

Chat soon,

Leonard

[Jason Short]

Also posted to GIT issues list –

When using Follow-me mode via the tablet, it continually sets Guided mode waypoints. This makes it impossible to recover control of the craft if the tablet remains in Follow-me.
Can I propose we only enter guided-mode from Loiter? If the user exits Guided/Loiter mode, the aircraft will refuse any further guided mode packets. Tablets will need to successfully enter Loiter, confirm the mode, then begin guided. Tablets should then watch for the mode change via mavlink to exit follow-me.

Jason

[Chris Anderson]

Just to be clear, you're referring to Droidplanner, right? Any idea if Andropilot exhibits the same behavior?

-c

Jason

--
You received this message because you are subscribed to the Google Groups "drones-discuss" group.
To unsubscribe from this group and stop receiving emails from it, send an email to drones-discus...@googlegroups.com.
For more options, visit https://groups.google.com/groups/opt_out.

--
Chris Anderson

CEO, 3D Robotics

ch...@3drobotics.com

[Kevin Hester]

Andropilot automatically abandons follow-me mode if it sees any mode change from the vehicle.  So just toggling a mode switch will exit follow-me.  (it has always watched for the mavlink messages and confirms any mode changes)

[Brandon Basso]

That's right, this is the main difference between followme in DP v. AP.  Guided mode is both a mode change and a waypoint, so on DP it continually updates the waypoint and resets the mode to guided, even if you switch out of guided mode with the handset.

Another difference: I couldn't find a way to set the guided mode altitude in DP, whereas this is a setting in AP. Where does guided mode get it's altitude from in general when you set it from a tablet? (Mission planner requests that you set the altitude when you switch to guided mode).

Brandon

[Kevin Hester]

also - I'd love feedback on lead-it in Andropilot: If you set the min/max follow-me distance to something other than the default of zero, it will try to hold the same relative bearing to the pilot (i.e. if you turn it should stay behind you/in front of you based on the magnetometer in the tablet).  If you set min/max follow me to _different_ numbers then tilting the tablet should cause the vehicle to move closer or farther - pointing as if with a mouse... (though testing coverage for this feature is low - so be a bit more careful)

[Dirk Milz]

Hey Pro´s!

Today I was able to test the autotune-thing. What should I say? Perfect? Never expected such a easy and nearly perfect tuning for my copter.

Little bit suprided about the high Stab values, but copter is flying like never before.

If anybody interested I attachted some logs for analysing.

Frame:

spider-like Frame, but symetrical.

Center of Gravity = center of Lift = center of APM2.5

X-Axis is high inertia because of Gimbal with Cam in Front and Battery(600g) in the Back.

TOW = 1880g

Thanks, thanks, thanks!

cheers

Dirk