plane: tail-sitter interpolate gains with airspeed and enable Q_Assist #9167

IamPete1 · 2018-08-09T21:18:57Z

Interpolate Gains between VTOL gains and forward flight gains in Q modes based on airspeed.

Do this based on two new parameters that define the airspeed range at which the change over occurs.

This is a significant improvement at high airspeed although is not perfect. This would benefit from a plane yaw rate controller and body frame rather than earth frame yaw control. Hope to add Q_assist support for this code at some-point so the plane modes can also benefit from this gain interpolation. Also a QAcro mode would be fun.

Significantly adds to the Q modes flight envelope for tail sitters.

Tested in SITL.

https://www.youtube.com/watch?v=4sVT3VI5HDw&feature=youtu.be

IamPete1 · 2018-08-12T12:35:52Z

some examples of the issue in real aircraft, not as obvious as you back off as soon as it happens
https://www.youtube.com/watch?v=ktcWGpBBqEQ&feature=youtu.be&t=9m54s
https://www.youtube.com/watch?v=ktcWGpBBqEQ&feature=youtu.be&t=18m38s
https://www.youtube.com/watch?v=0pJyAk8XwVQ&feature=youtu.be&t=1m17s

IamPete1 · 2018-08-12T23:11:37Z

I have reordered the code so it will work as intended for aircraft with forward flight thrust vectoring, And enabled Q_Assist, here is a demo of with Q_Assist and then without.
https://www.youtube.com/watch?v=6MIPEbqxRwE

robustini · 2018-08-13T11:34:23Z

Nice improvement!

IamPete1 · 2018-09-22T21:02:08Z

Have rebased, tidyed and squashed this

IamPete1 · 2018-10-06T15:47:31Z

@tridge , the model and params used in the realflight testing are here https://drive.google.com/open?id=17ZE85aQV1TUw1uq4aDS20j2ThS9BD-fI

Thanks

magicrub · 2019-02-03T15:51:05Z

ArduPlane/quadplane.cpp

        hal.util->get_soft_armed() &&
        ((plane.auto_throttle_mode && !plane.throttle_suppressed) ||
         plane.get_throttle_input()>0 ||
         plane.is_flying())) {
        // the quad should provide some assistance to the plane
-        if (transition_state != TRANSITION_AIRSPEED_WAIT) {


this change seems to have no effect, why change it?

to enable assisted flight for tail sitters without entering airspeed wait transition state

magicrub · 2019-02-03T15:52:05Z

ArduPlane/tailsitter.cpp

+        const float throttle = motors->get_throttle() * 100;
+        const float rud_gain = float(plane.g2.rudd_dt_gain) / 100;
+
+        // Calculate differential thrust in forward flight


why is all of this commented out?

originally it was there so it could be un-commented when plane got a yaw rate controller. I'm in the process of a rebase and tidy up.

magicrub · 2019-02-03T15:58:51Z

@kd0aij would you mind taking a look at this?

IamPete1 · 2019-02-03T19:03:08Z

this now includes #10381

I have rebased and tidyed now is working as it should, @magicrub would be great if you could have another look.

This significantly adds to the flight envelope of all modes for tailsitters. With one or two things that might catch you out. One is flying high alpha in FBWA, it works fine however the roll limit is for forward flight is still used so roll is much more twitchy than it would be at a equivalent angle in a Q mode. The second issue is going fast in Q modes, this locks you into the plane controller and it can sometimes get stuck in the plane controller attempting to reach a impossible angle. In both cases lots of throttle will help. I think both are also not helped by being at the extreme range of the euler angles for each mode. Also i suspect in real life you would be rather more cautious than i have been in SITL testing so may be very unlikely to happen at all.
TLDR: flight envelope is bigger but care must still be taken near the limits.

Plane modes don't really take advantage of the new high alpha capability because the don't know that at high alphas throttle is used for altitude control rather than pitch.

@kd0aij Would be great if you could take a look, I will probably add this to your copter tailsitter branch and have a play myself soon.

kd0aij · 2019-02-03T19:11:07Z

@IamPete1 I tried flying the AddictionX in RF8: OK in FBWA, but as soon as I switch to QHOVER, it crashes with backtrace reporting "no stack" in gdb. This is similar behavior to what I was getting with one of my recent PRs, but I never found the cause.
I checked out your Gain_interp branch: acb3b1b14752

IamPete1 · 2019-02-03T19:15:11Z

@kd0aij odd, working fine for me, the only thing that would cause it to crash is having the new parameters equal. This gives a divide by zero that is not checked... opoos i see i have set the defaults so they are equal.

IamPete1 · 2019-02-03T19:21:54Z

@kd0aij try that, have changed the defaults and added a divide by zero check.
As a rule of thumb you should tune Q modes for hover and then set the min scailing range to slightly lower than speed than the speed wobbles come on. And then set the max and Q_assist speed to FBW min speed or slightly above

ArduPlane/tailsitter.cpp

kd0aij · 2019-02-03T19:48:56Z

that fixed the crash, running qautotune on the AddictionX now...

Have you tried the AddictionX? After running qautotune, I have to set spdmin to zero to see an effect in qhover. spdmax is at 10

IamPete1 · 2019-02-09T17:12:31Z

@kd0aij I have just got round to trying the Addiction X, its has such big control surfaces and light weight it doesn't benefit much from this, if it had a bigger motor and could go faster i think you would start to see a difference. It benefits from adding rudder to the throttle scaling.

I have cherry picked your quad tail-sitter stuff on on top of this here https://github.com/IamPete1/ardupilot/tree/gain_interp_quad makes a huge difference on the striker in Q modes, can fly straight up at full throttle after a auto tune. Interestingly it failed to auto tune yaw, i think the removal of motor torque yawing means it did't finish, probably would have if i had lowered the aggressiveness. Again not such a benefit in forwards flight due to the delta wing already being good at high alpha, does kick on the throttles of the top and bottom motors tho.

kd0aij · 2019-03-05T01:38:48Z

@IamPete1 Do the PID controllers for plane use gains the same as the copter controller, i.e. does a gain of 0.5 result in the same rate loop gain for both? I assume that's true since you're interpolating between them, but I haven't looked at the plane controllers yet.

IamPete1 · 2019-03-05T09:40:51Z

@kd0aij no the gains don't directly scale. This runs both control loops and interpolates the outputs.

kd0aij · 2019-03-05T19:04:59Z

That makes sense. I was reading the PR title too literally, and didn't realize that both controllers are running at the same time. I guess the PID controller CPU cost is not significant.

…itters

kd0aij · 2019-03-10T17:59:13Z

ArduPlane/tailsitter.cpp

+
+    if (assisted_flight) {
+        control_stabilize();
+        motors_output(true);


We're calling motors_output more than once in each pass through set_servos, right?
The option to not run the rate controller is to make the multiple calls less inefficient?

this is for Q-assist, it we're in a normal vtol mode this is already done in the quadplane code, altough i'm not sure why I need to call control stabalise again

IamPete1 · 2019-03-10T19:22:07Z

@kd0aij I'm struggling to make any improvement on this by use of the yaw dammper. I'm beginning to think I'm asking too much for the airframe, to sort of knife edge and yaw up to vertical. Once its going fast full throttle on one motor and the other stopped is not enough to overpower the rudder not moving enough. My tvbs is quite unusual tho in that it actually has a rudder it might be fine for wings.

kd0aij · 2019-03-10T19:45:09Z

I don't see a yaw damper... is it part of this PR?

…

On Sun, Mar 10, 2019 at 1:22 PM Peter Hall ***@***.***> wrote: @kd0aij <https://github.com/kd0aij> I'm struggling to make any improvement on this by use of the yaw dammper. I'm beginning to think I'm asking too much of the airframe, to sort of knife edge and yaw up to vertical. Once its going fast full throttle, one motor on full and the other stopped is not enough to overpower the rudder not moving enough. My tvbs is quite unusual tho in that it actually has a rudder it might be fine for wings. — You are receiving this because you were mentioned. Reply to this email directly, view it on GitHub <#9167 (comment)>, or mute the thread <https://github.com/notifications/unsubscribe-auth/ACMZPQEuQECWRGejlolzJOsXyV22JODZks5vVVtigaJpZM4V2xrf> .

-- Mark Whitehorn kd0aij@gmail.com

IamPete1 · 2019-03-10T19:48:18Z

@kd0aij https://github.com/IamPete1/ardupilot/tree/Gain_intep_yaw I havent pushed it here yet

kd0aij · 2019-03-10T20:07:20Z

It's definitely an airframe limitation; you need to be able to do a knife-edge loop to recover using rudder only. Roman Bapst implemented a recovery technique (for PX4) which first rolls level, then pitches up. That can happen very fast... There's actually an academic paper on the technique, from ETH Zurich, I think.
This is one reason I'm exploring quad tailsitters, but I haven't managed a knife-edge loop with the Stryker_quad yet :( I think it should have enough lateral area to do it though.

IamPete1 · 2019-03-10T20:18:57Z

@kd0aij Thats good to know I guess, can you have a fly of this branch with these params https://github.com/ArduPilot/SITL_Models/blob/quad_TS/RealFlight/markw/Tailsitters/QuadMotorTailsitters/Stryker/matrixTS_ptun%20gain%20scailing.parm The Q gains are straight from auto tune, would be interested to hear what you think. (just did a knife edge loop in Qacro, not very tidy tho more of a oval)

peterbarker · 2020-12-04T23:45:31Z

@IamPete1 I'm guessing the tailsitter code has moved on significantly since this PR was created.

Do you see any value in leaving this PR open?

IamPete1 · 2020-12-05T00:14:39Z

@peterbarker I do have fairly recent rebase of this, as in this year, I do think this would be better, but hopefully #15719 will be close.

Not much chance of getting back to this unless some keen testers turn up, I will close.

IamPete1 mentioned this pull request Aug 12, 2018

Tailsitter: Improve scaling of control surfaces #8107

Closed

IamPete1 changed the title ~~Tail Sitter interpolate gains with airspeed~~ Tail Sitter interpolate gains with airspeed and enable Q_Assist Aug 12, 2018

khancyr added Plane VTOL-Plane and removed Plane labels Aug 13, 2018

IamPete1 force-pushed the Gain_interp branch 2 times, most recently from 206f9de to 2affa86 Compare September 22, 2018 19:48

OXINARF requested a review from tridge September 24, 2018 17:48

OXINARF added the DevCallTopic label Sep 24, 2018

IamPete1 force-pushed the Gain_interp branch 4 times, most recently from d38f6b7 to ca27d31 Compare September 25, 2018 17:14

tridge self-assigned this Oct 1, 2018

CraigElder removed the DevCallTopic label Nov 12, 2018

IamPete1 mentioned this pull request Jan 20, 2019

Plane: allow copter tailsitters kd0aij/ardupilot#5

Closed

IamPete1 force-pushed the Gain_interp branch from ca27d31 to fefd38f Compare February 3, 2019 15:25

magicrub reviewed Feb 3, 2019

View reviewed changes

IamPete1 added the WIP label Feb 3, 2019

IamPete1 force-pushed the Gain_interp branch from 87f1626 to acb3b1b Compare February 3, 2019 18:49

IamPete1 changed the title ~~Tail Sitter interpolate gains with airspeed and enable Q_Assist~~ plane: tail-sitter interpolate gains with airspeed and enable Q_Assist Feb 3, 2019

IamPete1 removed the WIP label Feb 3, 2019

kd0aij reviewed Feb 3, 2019

View reviewed changes

ArduPlane/tailsitter.cpp Outdated Show resolved Hide resolved

kd0aij mentioned this pull request Mar 4, 2019

tailsitter gain scaling with support for airspeed sensor for high-speed flight in Qmodes #10690

Merged

IamPete1 added 2 commits March 9, 2019 20:52

plane: add params for gain interpolation and enable Qassist for tails…

8c481c5

…itters

plane: tail sitter add gain interpolation with airspeed

6ae8484

IamPete1 force-pushed the Gain_interp branch from f65065e to 6ae8484 Compare March 9, 2019 20:52

kd0aij mentioned this pull request Mar 10, 2019

iampete interp gains #10747

Closed

kd0aij reviewed Mar 10, 2019

View reviewed changes

IamPete1 closed this Dec 5, 2020

Provide feedback

Saved searches

Use saved searches to filter your results more quickly

plane: tail-sitter interpolate gains with airspeed and enable Q_Assist #9167

plane: tail-sitter interpolate gains with airspeed and enable Q_Assist #9167

IamPete1 commented Aug 9, 2018

IamPete1 commented Aug 12, 2018

IamPete1 commented Aug 12, 2018

robustini commented Aug 13, 2018

IamPete1 commented Sep 22, 2018

IamPete1 commented Oct 6, 2018 •

edited

Loading

magicrub Feb 3, 2019

IamPete1 Feb 3, 2019

magicrub Feb 3, 2019

IamPete1 Feb 3, 2019

magicrub commented Feb 3, 2019

IamPete1 commented Feb 3, 2019

kd0aij commented Feb 3, 2019 •

edited

Loading

IamPete1 commented Feb 3, 2019

IamPete1 commented Feb 3, 2019

kd0aij commented Feb 3, 2019 •

edited

Loading

IamPete1 commented Feb 9, 2019 •

edited

Loading

kd0aij commented Mar 5, 2019

IamPete1 commented Mar 5, 2019

kd0aij commented Mar 5, 2019

kd0aij Mar 10, 2019

IamPete1 Mar 10, 2019

IamPete1 commented Mar 10, 2019 •

edited

Loading

kd0aij commented Mar 10, 2019 via email

IamPete1 commented Mar 10, 2019

kd0aij commented Mar 10, 2019

IamPete1 commented Mar 10, 2019 •

edited

Loading

peterbarker commented Dec 4, 2020

IamPete1 commented Dec 5, 2020

plane: tail-sitter interpolate gains with airspeed and enable Q_Assist #9167

plane: tail-sitter interpolate gains with airspeed and enable Q_Assist #9167

Conversation

IamPete1 commented Aug 9, 2018

IamPete1 commented Aug 12, 2018

IamPete1 commented Aug 12, 2018

robustini commented Aug 13, 2018

IamPete1 commented Sep 22, 2018

IamPete1 commented Oct 6, 2018 • edited Loading

magicrub Feb 3, 2019

Choose a reason for hiding this comment

IamPete1 Feb 3, 2019

Choose a reason for hiding this comment

magicrub Feb 3, 2019

Choose a reason for hiding this comment

IamPete1 Feb 3, 2019

Choose a reason for hiding this comment

magicrub commented Feb 3, 2019

IamPete1 commented Feb 3, 2019

kd0aij commented Feb 3, 2019 • edited Loading

IamPete1 commented Feb 3, 2019

IamPete1 commented Feb 3, 2019

kd0aij commented Feb 3, 2019 • edited Loading

IamPete1 commented Feb 9, 2019 • edited Loading

kd0aij commented Mar 5, 2019

IamPete1 commented Mar 5, 2019

kd0aij commented Mar 5, 2019

kd0aij Mar 10, 2019

Choose a reason for hiding this comment

IamPete1 Mar 10, 2019

Choose a reason for hiding this comment

IamPete1 commented Mar 10, 2019 • edited Loading

kd0aij commented Mar 10, 2019 via email

IamPete1 commented Mar 10, 2019

kd0aij commented Mar 10, 2019

IamPete1 commented Mar 10, 2019 • edited Loading

peterbarker commented Dec 4, 2020

IamPete1 commented Dec 5, 2020

IamPete1 commented Oct 6, 2018 •

edited

Loading

kd0aij commented Feb 3, 2019 •

edited

Loading

kd0aij commented Feb 3, 2019 •

edited

Loading

IamPete1 commented Feb 9, 2019 •

edited

Loading

IamPete1 commented Mar 10, 2019 •

edited

Loading

IamPete1 commented Mar 10, 2019 •

edited

Loading