AP_AHRS: prevent DCM fallback when losing GPS

[robertlong13]

DCM is particularly bad during hover modes and should only be used when the EKF cannot safely do its job. Currently, losing the GPS during VTOL takeoff triggers an unnecessary DCM fallback.

This PR changes the const_pos_mode flag check so that it only triggers fallback if that happens despite having GPS.

[rmackay9]

This is probably OK but we should be sure that there is no change in the Copter's reported position when we gain/lose GPS. For extra points we should test the behaviour when using a non-GPS sensor for position estimation (e.g. visual odometry/vicon).

[robertlong13]

This is probably OK but we should be sure that there is no change in the Copter's reported position when we gain/lose GPS. For extra points we should test the behaviour when using a non-GPS sensor for position estimation (e.g. visual odometry/vicon).

You mean the Plane's reported position? This is all under an if block than only runs for Plane and Rover.

I'll see about testing these cases, but I need to learn about them and how to sim them first.

[robertlong13]

Thinking about this more, you're right that I probably should be checking if it has any kind of device that can provide AID_RELATIVE or AID_ABSOLUTE is active, not just GPS (i.e., something should be able to keep the EKF from going const_pos_mode, but something is going very wrong, meaning we need the fallback).

[IamPete1]

I wounder if it should also check fly_forward, that would restrict this change of behavior to only VTOL modes on plane.

[robertlong13]

I wounder if it should also check fly_forward, that would restrict this change of behavior to only VTOL modes on plane.

For some reason, in my testing (though that's been limited), killing GPS only seems to trigger DCM fallback in hover. Haven't tested on a real bird in forward flight though.

I'll give the fly_forward check some thought though. If I do that, I'd rather go the extra step of fully disallowing DCM in hover, like Copter does always.

[timtuxworth]

During my no-GPS testing with a fixed-wing - mostly flying fixed wing AUTO mode, the plane would fallback to DCM within a few seconds of me turning off the GPS - it was very annoying because I was trying to test EKF dead-reckoning.
[Not a VTOL].

[IamPete1]

For some reason, in my testing (though that's been limited), killing GPS only seems to trigger DCM fallback in hover. Haven't tested on a real bird in forward flight though.

That is because EKF does not throw its toys out of the pram if flyforward is true and it looses GPS. (provided everything else is good) So as you say adding that check should not change the current behavior.

Everything we can do to minimize the scope of the change means less testing and fewer edge cases.

[Ryanf55]

Tridge recommends this approach because it's vehicle-dependent:
#25676

[robertlong13]

I like Tridge's approach better. Closing