improved implementation of tailsitter bodyframe-roll control

[kd0aij]

graphical visualizations of the current implementation showed unwanted coupling of pitch and yaw to roll:
https://www.glowscript.org/#/user/kd0aij/folder/QuadPlane/program/bodyframe-master

updated version makes bodyframe-roll independent of pitch and yaw:
https://www.glowscript.org/#/user/kd0aij/folder/QuadPlane/program/bodyframe-roll-update

Also reduced the yaw error threshold from 90 to 30 degrees to reduce dynamic yaw (heading) errors at the expense of yaw rate tracking.

Flight test by @losawing: https://discuss.ardupilot.org/t/dual-motor-tailsitters/15302/2496

[IamPete1]

Looks OK. This is your input function that is not widely used, I think this has been tested by losawing? Basically if your happy with it then I'm happy for it to be merged.

[robustini]

No news about that?

[lthall]

I can't review this PR.

[kd0aij]

@lthall Thanks for responding; I assume you don't have a problem with it being merged.

[lthall]

Sorry, that statement was too short. I can't review this PR because I don't understand the original code design, so I can't evaluate the effectiveness of the changes.

While I often get asked to review code I don't fully understand. I do so if there are parts of the code that are in areas I am responsible for and I feel the other reviewers can cover the areas I do not feel I can review effectively.

I also don't want to stifle the development of other aircraft and in this case there are elements in the control structure that I do not understand and the comments don't explain. So I can't provide any value with my review, for or against.

[kd0aij]

I had hoped that the VPython implementation and visualizations (linked in the first comment) would clearly show the intent of the code and the differences between the two implementations.

But it's also clear that the quaternion manipulations in function bodyFrameRoll_eulerPitchYaw() are very difficult to understand hence probably more complicated than necessary :)

[robustini]

Tested today in reality and in QStabilize I had an unpleasant behavior.
Watch the behavior of the pitch.

2022-09-11 10-54-17.zip

[kd0aij]

@robustini Thanks for testing.
Because you had parameter Q_TAILSIT_INPUT=0, bodyframe roll was not selected, and this PR had no effect.
To test this PR, you would need to set bit 1 of that parameter, so 2 instead of 0 for copter mode and 3 for plane mode.
I assume the issue is the failure to track desired pitch here?

[robustini]

@kd0aij Damn Mark, you're right, thanks!
So I have to find out why she made that joke, also because I always flew with those parameters but I haven't used the master for two months.