Yaw Drift due to increase of EKF_GYRO_BIAS estimation

Blue Robotics Software ArduSub
1

I have Pixhawk 2.4.8 otopilot with ArduSub_V3.5. 4 firmware installed on a custom ROV.

The problem (figure below) is that yaw (ATT.Yaw) info displayed on QGroundControl_v3.5.3 switches between EKF primary (NKF1.Yaw) and secondary (NKF6.Yaw) solutions, when the vehicle is stationary. Also, EKF secondary yaw solution (NKF6.Yaw) drifts due to drift in gyro bias estimation (NKF6.GZ).

log_0_2019-5-20-15-12-06_Yaw_Switching_and_Secondary_Drifting
log_0_2019-5-20-15-12-06_Yaw_Switching_and_Secondary_Drifting1705×987 131 KB

I have checked that Magnetometers are well calibrated and give correct results.

log_0_2019-5-20-15-12-06_Magnetometers_are_fime
log_0_2019-5-20-15-12-06_Magnetometers_are_fime1705×987 141 KB

Although Gyro-Z measurement from first IMU (IMU1.GyrZ) seems stationary, IMU2 (IMU2.GyrZ) seems to be drifting, but I dont thing this is a problem because the values are very small and I dont see and problems in other tests.

log_0_2019-5-20-15-12-06_IMU_GYRO_IMU2GyrZ_drifting
log_0_2019-5-20-15-12-06_IMU_GYRO_IMU2GyrZ_drifting1705×987 199 KB

When I rotate the vehicle at 90deg intervals, I see that yaw values (NKF1.Yaw and NKF6.Yaw) are drifting even though the vehicle stays stationary. NKF1.GZ and NKF6.GZ values shows huge changes.

log_0_2019_-5-23-09-10-12_90deg_Test_Yaw_Drifting
log_0_2019_-5-23-09-10-12_90deg_Test_Yaw_Drifting1705×987 121 KB

I have checked that Magnetometers are well calibrated and give correct results.

log_0_2019_-5-23-09-10-12_90deg_Test_Magnetometers_are_fine
log_0_2019_-5-23-09-10-12_90deg_Test_Magnetometers_are_fine1705×987 127 KB

Gyro measurements seems fine…

log_0_2019_-5-23-09-10-12_90deg_Test_Gyros_are_fine
log_0_2019_-5-23-09-10-12_90deg_Test_Gyros_are_fine1705×987 133 KB

I think the problem is that Gyro.Z bias estimations in EKF do grows too much to make Yaw estimation drifting. Magnetometer and Gyro sensors seems to be measuring correctly. Do you have any suggestions?

Best Wishes.

2

Hi, I’ve reported similar issue here: Slow rotation around z axis even staying on a table - #2 by patrickelectric
Have you found a solution to that problem?

BR,
Alex

3

Hi @slovak194,

Although I have experimented with different EKF settings and external magnetometers, I couldn’t reach a solution so far, but I believe that EKF settings need to be calibrated for ROVS having metals and noisy electronics. The nature of yaw drift is very different when the ROV is in or out of water. The interesting thing is that I don’t see any problems with roll and pitch, but only a yaw drift.

I am still waiting for firmware update for ArduSub, and expecting the capabilities implemented in ArduPlane v4.0.1 and newer versions. I am sure they have done outstanding contributions to EKF firmware since ArduSub v3.5.4.

@patrickelectric do you have scheduled plan for releasing an updated firmware since you moved to ArduPilot codebase, so that we can benefit from new features.

Best Regards.

4

We are wrapping a 4.0 release now. The 4.0 beta is available for download at firmware.ardupilot.org and in QGC if you use ‘advanced settings’.

5

The nature of yaw drift is very different when the ROV is in or out of water.

I also see that. I guess that this can be caused by different temperature conditions. As robot is cooled much better underwater, especially in winter.

BR, Alex

6

That is great news @jwalser. Can’t wait to test new features.

7

Is there a solution on the problem? We have a similar problem.

We have testet sevral of pixhawk, same drifting problem on yaw.

Navio2 seems to have more stable yaw adjustment.

8

I still experience the same problems.