EKF3 IMU core unhealthy, stopped aiding (esp. after landing)

ArduCopter Copter 4.1
1

Hello –
I am running into this series of errors that prevent GPS-based simple PosHold on 4.1/4.2-dev. The parameter settings are meant to enable GPS as the primary position source for XY and Z (this is a ZED F9P RTK board in rover mode [not moving base]). I consistently see messages such as “core unhealthy” and “IMU stopped aiding”.

In the attached log file,

  • I try arming in PosHold, and it doesn’t arm (“Need position estimate”), even though the RTK module is in Fix mode.
  • I can arm and take-off in Stabilize and the second attempt to switch to PosHold works, and the vehicle holds position.
  • Once I land, this switch to PosHold is never successful.

I’ve checked the logs for Vibrations (foam-mounted MPU6000), and those look ok. The only thing is that the innovations in EKF PD and VD are sometimes excessive - no idea why.

Any help/directions welcome! This has been a problem for a while now. I have previously used Pixhawk 2.4.8s and CubeBlacks without issues on 4.0dev.

Current setup: 5" props, DShot150, w/ Kakute F7, external I2C compass. The ZED-F9Ps get
good RTK-Fix.

Any ideas on what might be off? Hardware/software?
Thank you!

2

Here is some more on this.

It appears that EKF innovations are all over the place, especially after a landing. A mode change to POSHOLD will always fail afterwards, even though the GPS status is always 6 (= RTK FIX).

Here, the Baro altitude and GPS altitude (shifted) look consistent, while GPS status is always at 6:

alt-gps_baro
alt-gps_baro1673×1052 23.8 KB

And yet, IPD spikes upon landing (pink), while IVN and IVE (blue and yellow) run off quite freely:

innov-3d
innov-3d1698×1077 22.2 KB

This causes EKFCHECK-2 and “IMU stopped aiding” at landing, and never again does POSHOLD succceed. Here’s a zoomed-in region of the first time this happens:
innov3d-zoomed
innov3d-zoomed1698×1073 19 KB

Note that GPS status is still 6 throughout, and POS log contains lat/lon/alt as expected. VIBE looks normal (the IMU is vibration damped).

Any pointers from the devs, maybe? Here’s this log file.