EKF2 yaw estimation error

[totomyl]

Hello,

I am trying to send motion capture data to a pixfalcon running px4 autopilot version 1.8.2. The motion capture system (mocap optitrack) is being published to the topic /mavros/vision_pose/pose. I am currently using EKF2 to estimate the position of the drone, I have set EKF2_AID_MASK to 24 (yaw + vision fusion). The barometer is disabled as well as the magnetometer and gps.

When visualizing the data in RVIZ, I am able to see both the mavros/vision_pose/pose and mavros/local_position/pose topics, however, the local_position topic keeps on yawing while the drone is still. When the drone is being moved around, both the local_position and the vision_pose move but the local_pose is constantly yawing so they are never in line.

I have tried upgrading the px4 firmware to 1.9.2 but this did not solve the problem. The sensors were re-calibrated multiple times and that did not help either.

Here is the log https://review.px4.io/plot_app?log=a72dd458-dc4f-4650-8335-02242da6e094
link to download: https://drive.google.com/open?id=16aOQq7O2m_1sBSl38vRLrrPXAQSijeY9

Any help or guidance would be appreciated, thank you in advance!

[RomanBapst]

@totomyl You will need to provide a log otherwise we can't help you.

[totomyl]

@RomanBapst sorry about that, I have added the link to the flight review, I was not sure how to put the actual file onto here as it was not supported.

[artifactz]

Maybe you can take a look at rostopich echo /mavros/imu/data_raw to check if the yawing already happens there. Also, your EKF2_MAG_NOISE parameter seems very high (EKF2_EV_GATE as well).

[julianoes]

@totomyl is this a duplicate of #12577 or something else?

[totomyl]

@totomyl is this a duplicate of #12577 or something else?

#12577 is a different problem, the mag won't even disable on it

[stale]

This issue has been automatically marked as stale because it has not had recent activity. Thank you for your contributions.

[bresch]

This was caused by an issue in the gyro bias estimate (due to some numerical issues in the covariance prediction step). It's now fixed on main and release 1.13