Feature/TPA #5861

anton-matosov · 2016-11-15T09:25:44Z

Re-implement Throttle PID Attenuation to support per-component setting of attenuation break and slope rate.
Used a bit different function that has preductable curve based on inputs.
You can compare both functions here https://www.desmos.com/calculator/gn4mfoddje

(PR includes new ZMR250 frame rc and mix files)

…ive function

LorenzMeier · 2016-11-15T09:33:21Z

Awesome! Do you have some feedback on MC_RATE_I_MODE?

anton-matosov · 2016-11-15T09:40:16Z

I have removed it from this PR, I am still trying to figure out how this code affects flight characteristics

kd0aij · 2016-11-15T16:41:22Z

@anton-matosov Interesting; have you done any testing to show the motivation for and effectiveness of also scaling the I and D gains? Also, what is more predictable about the scaling now?

anton-matosov · 2016-11-15T18:04:07Z

My drone had D oscilations at 0.8+ throttle that I couldn't compensate with existing TPA. That was the main motivation for these changes.
While implementing it I have found that it's hard to predict what will be your PID rate attenuation with max throttle for particular break and slope values.
With my function you always have (slope at full throttle==PID scale factor) and linear function from breakpoint to full for the scale.
You can see graphs for both functions here https://www.desmos.com/calculator/gn4mfoddje and play with b (break) and r (rate/slope) sliders

anton-matosov · 2016-11-15T18:39:20Z

src/modules/mc_att_control/mc_att_control_main.cpp

@@ -1005,6 +1053,63 @@ MulticopterAttitudeControl::task_main()
 					_controller_status_pub = orb_advertise(ORB_ID(mc_att_ctrl_status), &_controller_status);
 				}
 			}
+
+			if (_v_control_mode.flag_control_termination_enabled) {


This change should be in my RC failsafe PR #5863

kd0aij · 2016-11-16T00:41:55Z

@anton-matosov Thanks,that makes sense. I hadn't noticed a need for decreasing D when tuning and my procedure was just to increase the slope till the oscillations were inaudible. Hence I didn't really care what the actual attenuation at max throttle was, though I did calculate it afterward. With the new scaling, both r and b affect the slope, so it might be best to rename the r parameter to something which more accurately names its function.

anton-matosov · 2016-11-16T01:46:01Z

Agreed. In other flight stacks it's called TPA Rate
I will rename it tonight

LorenzMeier · 2016-11-19T13:58:46Z

@anton-matosov This looks awesome. Is it feasible to enable a safe default for all platforms? Do you have a chance to test with a 450 sized drone?

anton-matosov · 2016-11-19T15:22:48Z

Unfortunately I have only 250 racer and 550 hexa (which weight 2 kg). I was ok to test full throttle with racer, but not comfortable at all to do this with hexa.
IMO default for TPA - is no TPA, as usually PIDs are not tuned to their extremes so there are no oscillations at higher/full throttle.

# Conflicts: # src/modules/mc_att_control/mc_att_control_params.c

anton-matosov · 2016-12-01T05:48:28Z

Can anyone please take a look at this PR?

bkueng

Can you change the autostart id 4070 to something else? This conflicts with the recently merged Intel Aero board.

anton-matosov · 2016-12-02T18:11:09Z

Sure, will change to 4080

anton-matosov · 2016-12-10T02:13:17Z

@kd0aij P gains don't even participate in the acro mode, only rates. The main influencer of the attitude hold for acro mode is I rate, I have found it should be in the range of 0.19-0.25 for my frame/motors/esc/battery/props setup.

@LorenzMeier parameters are actually very unique for frame/motors/esc/battery/props combination. If I simply change props I already need to re-tweak parameters (deviation depends on how big is the change).
ZMR250 and QAV250 frames are very similar, but other components are not.
In addition to the frame size we should consider combinations of other components:

battery 3s vs 4s (totally different P and D rates) + vendor (different discharge rate and curve which affects TPA settings)
esc (different refresh frequency and motor timings and firmware, can drastically affect final thrust and require different P and D rates)
motor (generally only KV matters, totally different P and D rates for 2050kv and 2400kv)
props (diameter and pitch affect tuning significantly, e.g. 5030 2 blade vs 5051 3 blade requires different P and D rates)
final weight also matter

Setup for which zmr250 config was created is as follows:

battery: 2200mah 3s
motors: 1808 V-Spec 2400kv
esc: Afro 20A Race Spec ESC running SimonK
props: Gemfan 5030 2 blade

I will put this info into config as well.

I have switched to 3 blades recently and tried 2 different kinds:

For DAL 5045 quad become a bit sluggish, so I had to lower D a little bit
For RaceKraft RK5051 I also got oscillations on higher throttle and sharp turns, so I will have to adjust P rate and TPA settings (didn't have time to find perfect settings yet).

I have some flight videos captured for current settings and RK5051 props. Once I get a change to process them I will share them in the px4 discussion forums.

kd0aij · 2016-12-10T04:09:48Z

@anton-matosov Of course I was referring to the rate P gains; sorry for the ambiguity in my comment.
The roll/pitchrate_p gains for the ZMR250 in this PR are .05, whereas those for the generic 250 are .19

anton-matosov · 2016-12-10T04:12:11Z

Yep, that is what I am flying at, and with RK5051 props I will have to lower them even more

anton-matosov · 2016-12-10T04:12:25Z

Here is one of the FPV flights https://youtu.be/-SNok411Btc
and other one https://youtu.be/dqQlU0KXfS8

kd0aij · 2016-12-10T04:22:33Z

Thanks for posting video; that helps a lot for comparison. I'm running 3S, 2300KV motors and 5.4.6 Hyperion bullnose 2-blade props at a field elevation of 1760m. Hard to believe the altitude makes such a big difference; I wonder if something else is different?

I'd appreciate it if you could post logs using the new logger, with lots of rapid roll/pitch action. (easier to do in stabilize mode without crashing) That would let me run an impulse response analysis for comparison with my racers.

BTW, the current TPA eliminated high-throttle oscillations for me without degrading overall performance.

anton-matosov · 2016-12-10T04:25:21Z

You are welcome.
Sure I can do that. How to enable new logger? (I have never used it before)

kd0aij · 2016-12-10T04:28:09Z

Assuming you're using Pixracer, all you have to do is use QGC to set parameter SYS_LOGGER to "new logger".

anton-matosov · 2016-12-10T04:29:26Z

I have just double checked the config and it seems to be wrong.
I will update it now
Never mind, looked at the wrong commit.

kd0aij · 2016-12-10T04:38:36Z

Why have you reduced the multirotor mixer roll_scale parameter for the zmr250? Are you using that to compensate for a lower moment of inertia on the roll axis? My approach would be to lower the rate_P gain for the roll axis instead.

anton-matosov · 2016-12-10T04:45:15Z

Are you referring to MC_PITCH_TC and MC_ROLL_TC? these parameters work as simple PID fine tune scale factor (MC_ROLLRATE_P/I/D * 0.2/MC_ROLL_TC, same for pitch). There is no real need to change them, you can simply adjust your PIDs, but I have changed them during original tuning and got current PIDs with TC applied and it was flying really well so I decided to leave them.
These TC params are used by QGroundControl for "roll/pitch sensitivity". I have confused them with RC rates (which are actually "encrypted" in MC_ACRO_R/P/Y_MAX params).

In the next iteration of tuning for 3 blades I am going to remove MC_*_TC from config

LorenzMeier · 2016-12-10T10:42:01Z

Thanks for the good details @anton-matosov, it makes a lot more sense now. It would be great if you could work with @DonLakeFlyer on a tuning UI for multicopters that is more user-friendly (new or different sliders).

LorenzMeier · 2016-12-10T11:44:56Z

I've rebased this, removed the merge commit, removed all noisy (fix ABC) commits and was left with one clean commit. See master.

LorenzMeier · 2016-12-10T11:45:10Z

@anton-matosov Thanks for all the hard work!

kd0aij · 2016-12-10T14:07:10Z

ROMFS/px4fmu_common/mixers/zmr250.main.mix

@@ -0,0 +1,10 @@
+# R: <geometry> <roll scale> <pitch scale> <yaw scale> <deadband>
+R: 4x 7654 10000 10000 0


@anton-matosov I'm talking about the roll_scale parameter (7654) on this line

Ok, never mind my previous post then.
Mixer config is the best way to accommodate custom geometry. ZMR 250 is asymmetric and mixer compensates this.
For ZMR250 mixing for cleanflight/betaflight please refer to https://oscarliang.com/custom-motor-output-mix-quadcopter/

anton-matosov · 2016-12-15T07:16:51Z

@LorenzMeier
I have performed some testing of the acro mode on SK450 frame with 11kv motors, 10x4.5 slow-fly props and 3s battery. And it doesn't seem to have any noticeable oscillations on full throttle punches.
I have not finished PID tuning tho (broke frame), but I have noticed that I can/need get much higher D rate gain compared to racer setup before I get response without overshoot and even much higher gain to get sluggish response or D style twitchiness.
So behavior of PID loop in acro for bigger frame is very different.

kd0aij · 2016-12-15T14:03:12Z

I'd like to see flight logs for both your 250 and 450 quads (in the new .ulg format) for impulse response analysis.

…

On Thu, Dec 15, 2016 at 12:16 AM Anton Matosov ***@***.***> wrote: I have performed some testing of the acro mode on SK450 frame with 11kv motors, 10x4.5 slow-fly props and 3s battery. And it doesn't seem to have any noticeable oscillations on full throttle punches. I have not finished PID tuning tho (broke frame), but I have noticed that I can/need get much higher D rate gain compared to racer setup before I get response without overshoot and even much higher gain to get sluggish response or D style twitchiness. So behavior of PID loop in acro for bigger frame is very different. — You are receiving this because you were mentioned. Reply to this email directly, view it on GitHub <#5861 (comment)>, or mute the thread <https://github.com/notifications/unsubscribe-auth/ACMZPS4ZhURs7dlKWe7bqg41S5jTXyD_ks5rIOllgaJpZM4KyTM4> .

anton-matosov · 2016-12-15T18:33:47Z

@kd0aij my 450 is flying pixhawk, not pixracer will new logger work there?

kd0aij · 2016-12-15T18:44:13Z

If you're building fmu-v2_default, you have to change the cmake config for that target to build logger instead of sdlog2. Then just set parameter SYS_LOGGER to the new logger and it should start up on boot.

anton-matosov · 2016-12-18T05:35:58Z

@kd0aij here are some flips and rolls logs for my ZMR250 (I hope this is the right file, FlightPlot doesn't plot any of it for some reason)
ZMR250 flips and rolls logs.zip
Unfortunately I broke my SK450 before I got a chance to grab logs for flips and rolls. So will get them for you once I fix it.

kd0aij · 2016-12-18T13:30:47Z

Thanks, Anton. This log looks fine in FlightPlot; make sure you have the latest version of PX4/FlightPlot installed. I'll run the impulse analysis today for comparison with my own 250s. Should be interesting. Here's a 450 size frame that would be hard to break: http://www.aquacopters.com/store/p2/Project_X-Frame_Only.html but so far I've only crashed mine into water. That saves a lot of props; I've only managed to break one prop that way.

…

On Sat, Dec 17, 2016 at 10:36 PM Anton Matosov ***@***.***> wrote: @kd0aij <https://github.com/kd0aij> here are some flips and rolls logs for my ZMR250 (I hope this is the right file, FlightPlot doesn't plot any of it for some reason) ZMR250 flips and rolls logs.zip <https://github.com/PX4/Firmware/files/659270/ZMR250.flips.and.rolls.logs.zip> Unfortunately I broke my SK450 before I got a chance to grab logs for flips and rolls. So will get them for you once I fix it. — You are receiving this because you were mentioned. Reply to this email directly, view it on GitHub <#5861 (comment)>, or mute the thread <https://github.com/notifications/unsubscribe-auth/ACMZPcbB-svKAvIIKfeF8P4yaTJDnb2Aks5rJMZAgaJpZM4KyTM4> .

kd0aij · 2016-12-18T14:52:55Z

@anton-matosov Your tuning looks great! Here's a link to the plots: https://github.com/kd0aij/impulse-response/tree/master/sample_ulogs/ZMR250/flips_250/plots
The rate impulse response is very close to critically damped for both roll and pitch, and the accel tracking has a latency of only 30msec.

anton-matosov · 2016-12-19T06:39:34Z

Thanks. I am very pleased to hear that I did it right 😊

anton-matosov · 2016-12-19T06:47:46Z

The waterproof frame looks interesting. How do you recover your quad from the water? What if it flips over?

kd0aij · 2016-12-19T13:57:39Z

A canoe is a good idea... Even with a receiver antenna taped to the inside bottom of the hull, I've not managed to flip back upright. Seems I lose the RC signal. But I'm sure I've seen videos of copters flipping back upright; would probably work with external whip antennas. A boat (or maybe a Golden Retriever) would still be a good idea though.

kd0aij · 2016-12-19T14:41:51Z

BTW, this is my S250 (3S) tuning: https://github.com/kd0aij/impulse-response/blob/master/sample_ulogs/S250AQ/2016-12-13/plots/rate_impulse.png

A little faster than yours, but a bit under-damped.

anton-matosov added 2 commits November 15, 2016 01:02

Rework TPA to have per-component setup and use more stable and intuit…

9a08b77

…ive function

Add ZMR250 params

40a70f9

Remove debug code for RATE_I_MODE

c4285a0

Fix formatting

195b618

anton-matosov commented Nov 15, 2016

View reviewed changes

anton-matosov added 4 commits November 15, 2016 20:25

Remove code that should be in MC-Failsafe branch

3a1c689

Rename TPA_SLOPE into TPA_RATE and remove unused params

26cf17f

Merge latest 'master' into feature/TPA

a4a795c

Merge branch 'master' into feature/TPA

e127e4a

LorenzMeier added this to the Release v1.5.0 milestone Nov 19, 2016

julianoes modified the milestones: Release v1.5.1, Release v1.6.0 Nov 22, 2016

Merge commit 'b020be13f67838047d1c2532cf66a99ceb28cddf' into feature/TPA

125cfb1

# Conflicts: # src/modules/mc_att_control/mc_att_control_params.c

Merge branch 'master' into feature/TPA

fb7eb50

bkueng requested changes Dec 2, 2016

View reviewed changes

anton-matosov added 3 commits December 2, 2016 10:12

Rename to solve conflict

7141b0f

Update to the most recent config

4df4226

Merge commit 'fee75c61a18d8c8d6a32406575ed5ede79d01446' into feature/TPA

50ff01d

Merge branch 'master' into feature/TPA

983b745

LorenzMeier closed this Dec 10, 2016

kd0aij reviewed Dec 10, 2016

View reviewed changes

anton-matosov deleted the feature/TPA branch December 15, 2016 07:17

jlecoeur mentioned this pull request Jul 15, 2019

PID rate controller - Add controller gain to support Ideal PID form (ISA standard) #12472

Merged

MaEtUgR mentioned this pull request Oct 16, 2019

Remove TPA from rate control #13208

Merged

		@@ -0,0 +1,10 @@
		# R: <geometry> <roll scale> <pitch scale> <yaw scale> <deadband>
		R: 4x 7654 10000 10000 0

Feature/TPA #5861

Feature/TPA #5861

Conversation

anton-matosov commented Nov 15, 2016 • edited Loading

LorenzMeier commented Nov 15, 2016

anton-matosov commented Nov 15, 2016

kd0aij commented Nov 15, 2016

anton-matosov commented Nov 15, 2016

anton-matosov Nov 15, 2016

Choose a reason for hiding this comment

kd0aij commented Nov 16, 2016

anton-matosov commented Nov 16, 2016

LorenzMeier commented Nov 19, 2016

anton-matosov commented Nov 19, 2016

anton-matosov commented Dec 1, 2016

bkueng left a comment

Choose a reason for hiding this comment

anton-matosov commented Dec 2, 2016

anton-matosov commented Dec 10, 2016

kd0aij commented Dec 10, 2016

anton-matosov commented Dec 10, 2016

anton-matosov commented Dec 10, 2016 • edited Loading

kd0aij commented Dec 10, 2016 • edited Loading

anton-matosov commented Dec 10, 2016

kd0aij commented Dec 10, 2016

anton-matosov commented Dec 10, 2016 • edited Loading

kd0aij commented Dec 10, 2016

anton-matosov commented Dec 10, 2016

LorenzMeier commented Dec 10, 2016

LorenzMeier commented Dec 10, 2016

LorenzMeier commented Dec 10, 2016

kd0aij Dec 10, 2016

Choose a reason for hiding this comment

anton-matosov Dec 10, 2016

Choose a reason for hiding this comment

anton-matosov commented Dec 15, 2016 • edited Loading

kd0aij commented Dec 15, 2016 via email

anton-matosov commented Dec 15, 2016

kd0aij commented Dec 15, 2016

anton-matosov commented Dec 18, 2016

kd0aij commented Dec 18, 2016 via email

kd0aij commented Dec 18, 2016

anton-matosov commented Dec 19, 2016

anton-matosov commented Dec 19, 2016

kd0aij commented Dec 19, 2016

kd0aij commented Dec 19, 2016

anton-matosov commented Nov 15, 2016 •

edited

Loading

anton-matosov commented Dec 10, 2016 •

edited

Loading

kd0aij commented Dec 10, 2016 •

edited

Loading

anton-matosov commented Dec 10, 2016 •

edited

Loading

anton-matosov commented Dec 15, 2016 •

edited

Loading