Plane: fast attitude recovery for quadplanes

[tridge]

This is a rework of #28320 which uses the fixed wing attitude controller for recovery instead of modifying the VTOL controller. This gives much faster attitude recovery and also avoids an issue with the way the quaternion controller may try to pitch down while inverted, leading to a large amount of altitude loss
This approach builds on the way we already use the fixed wing controller during forward transitions, so it is very little code change, and testing shows it produces a very small amount of altitude loss (eg. in the autotest the alt loss is usually around 2m from fully inverted)

[IamPete1]

This effectively recovers as it would in Q-Assist in a fixed wing flight. One issue I have seen on a real vehicle is that the fixed wing angle controllers have a bad time unusual attitudes. A vehicle got into a inverted spin and was no able to recover in Q-Assist. The issue was three fold, it was oscillating about roll +-180 and could not decide which way to go. The fixed-wing controllers hard limit angle error so the error was not building the longer it was upside down (as it would in the VTOL controller), so it never tried that hard to recover. The VTOL rate controller I terms are also relaxed in assist so again it was never trying that hard. In that case we were able to switch to a VTOL mode and have it recover quickly (although not quite quick enough to prevent the crash).

One advantage of this method is that we can say that it is not anything that the vehicle would not do in Q-Assist so were not in any more risk of overloading the vehicle that we would be in that case.

One disadvantage is that the user may be trying the VTOL mode because Q-Assist may have failed to recover, in which case it makes sense to try a different approach.

[tridge]

One issue I have seen on a real vehicle is that the fixed wing angle controllers have a bad time unusual attitudes. A vehicle got into a inverted spin and was no able to recover in Q-Assist. The issue was three fold, it was oscillating about roll +-180 and could not decide which way to go. The fixed-wing controllers hard limit angle error so the error was not building the longer it was upside down (as it would in the VTOL controller), so it never tried that hard to recover.

my preference is to apply this method with those limitations and work on improving the fixed wing attitude controller to better cope with these scenarios.

[tridge]

@IamPete1 I've added in the attitude control method as discussed, plus the indecision change for the FW roll controller
in the new test added in this PR we now lose no altitude at all (usually gains about 0.1 or 0.2 m in recovery). Without these changes we lose around 25m or so

[tridge]

need to do testing on some real quadplanes

[IamPete1]

I would like to see testing on a motor always on tailsitter like the swan H1 in realflight.

[tridge]

need to test the HWing in RealFlight

[tridge]

I would like to see testing on a motor always on tailsitter like the swan H1 in realflight.

I have tested in RealFlight. Flies fine. Inverted mode doesn't work, but also doesn't work in master. I also found a bug where if you disarm in a fwd flight mode in an all motor tailsitter the motors don't stop! That bug is in master

[tridge]

@IamPete1 I have been testing recovery using the Griffin model in RealFlight. I'm getting some strange results. One is that when I switch to QLOITER the aircraft doesn't slow down. For example:

log here: https://uav.tridgell.net/GriffinTest/00000042.BIN
zooming in, we see the pitch staying below 2 degrees until it suddenly goes to 5 degrees, then some time later it does to 30 degrees, which is what I expected to happen to slow down from 50 m/s.
Any idea what is limiting the pitch in QLOITER?

[tridge]

@IamPete1 this is another one that worries me:
https://uav.tridgell.net/GriffinTest/00000039.BIN
it dropped like a stone inverted

[IamPete1]

Any idea what is limiting the pitch in QLOITER?

We do have the transition pitch limit stuff, it works in all modes but is only likely to be noticeable in Q_LOITER, the key function is set_VTOL_roll_pitch_limit there is a time parameter Q_BACKTRANS_MS. If were still using the FW controllers we need to delay the init of loiter navigation and delay the start of that pitch up limit. We should be able to use the same code paths as tailsitters.

It does concern me for 4.6 that this is getting more and more complex.

[IamPete1]

I had another thought, we should update show_vtol_view to account for the delay in switching controllers so it is correctly reflected in the logs and GCS.

[tridge]

We do have the transition pitch limit stuff, it works in all modes but is only likely to be noticeable in Q_LOITER, the key function is set_VTOL_roll_pitch_limit there is a time parameter Q_BACKTRANS_MS

I looked at that, but it was set to 3000 ms, yet, the pitch was limited to under 2 degrees for more than 25 seconds.

I had another thought, we should update show_vtol_view to account for the delay in switching controllers so it is correctly reflected in the logs and GCS

yes, I have that change pending