EKF2 model not quite suitable for rover?

[lukegluke]

EKF2 model not quite suitable for rover?

[LorenzMeier]

There is no copter model in the estimator. You are feeding impossible physics to the estimator: If you have a step in position, then there must be a step in acceleration (but your log doesn't have that).

So the system correctly estimates a change in accelerometer bias (the only plausible explanation), which then leads to a correction of the pitch rates / gyro.

If you want to test an estimator, you need to feed it accurate physics, which means consistent position, velocity and acceleration.

[lukegluke]

You are feeding impossible physics to the estimator

Yes, I understand, it works fine in real case.
I just wonder and not quite understand how and why this change in accelerometer bias leads to angular corrections.
Anyway thanks for your participating.

[LorenzMeier]

I think for that you would need to read through the derivation of the state estimator. You could start here (more details in that folder): https://github.com/PX4/PX4-ECL/blob/master/EKF/documentation/Process%20and%20Observation%20Models.pdf