Self-Leveling Issue After Multiple Circles in Same Direction

Hi Guys,

We have a strange behaviour with V1.4.4

Description:
When doing multiple circles in the same direction ex: always to the left, the vehicule is not levelled anymore when releasing the sticks, then, we just have to switch back to POSITION mode, this forces it to stay in place, go back to Stabilize and it gets back to normal.

Issue starts at 2min15sec in the video when I release the stick and stop doing circles
at
2min45sec put it back to POSITION flight mode
at
2min55sec put it back to stabilize flight mode and it is levelled and doesn’t drift anymore.

Here is the LOG FILE
http://logs.uaventure.com/view/ZdaJxmt5tfbKv2p6BJe7nR

Using Stable 1.4.4
QGC MAC 3.0.1

Here is a video of it:

Regards,

pturcotte

You will want to switch the estimator to EKF2 (in the advanced param settings, under system params) and try again. The current default attitude estimator is suboptimal in that regard.

I tried it this morning and got the following results:

1- Stabilize mode works correctly,
2- Position mode… Forget that, it becomes very unstable. Do I have to recalibrate everything? Came back to local estimator and everything came back correctly.

thanks for your help,

Regards,

pturcotte

Can you please share a log file? Without that its not really possible to judge anything from your feedback.

I have recalibrated all the sensors and it works now. It is only loosing a lot of altitude when switching from Stabilize to POSITION, I have to sort this out, but it is perfectly stable now :slight_smile:

Thanks for your help !!!

Regards,

pturcotte

Hi Guys,

We finally managed to get a perfect setup with the F450 using the EKF2 position estimator.

It literally stays in place in POSITION mode and everything works good so far using V1.4.4

Thanks a lot guys !!!

Regards

pturcotte

@pturcotte I’d appreciate it if you would flight test this PR: https://github.com/PX4/Firmware/pull/5275
I have demonstrated that change to improve attitude estimation accuracy both in bench testing (on a turntable at 45RPM) and in flight (in a different branch). It would be nice to have your observations on how much it improves performance of the LPE/Q attitude estimator option.