Altitude performance on 1.8 sometimes worse than on 1.7

[AuterionWrikeBot]

Idea of Paul:
back to back testing on same hardware. Run 1.8 estimator on 1.7 and 1.8

[AuterionWrikeBot]

➤ Thomas Gubler commented:

@roman Bapst could you please provide the logs to paul?

[AuterionWrikeBot]

➤ Paul Riseborough commented:

Relates to #9740

[jfcisneros]

The following logs were obtained with a similar vehicle to the one referred in #9740 (comment) .
Altitude hold
v1.7.3
https://logs.px4.io/plot_app?log=f8cba56b-6eda-4ec0-8722-6dabc823f4c8
v1.8.0
https://logs.px4.io/plot_app?log=c4fddaa4-ef0f-4afb-bcb4-f9af3fcd396a
Position Control (No commands)
v1.7.3
https://logs.px4.io/plot_app?log=169ba77f-b011-4906-ad21-99a231776ab6
v1.8.0
https://logs.px4.io/plot_app?log=000b2a95-ef3e-40fe-8d82-4fcec21e7121

[jfcisneros]

Position Control (Position Setpoint in Offboard Mode)
Here is more noticeable the difference in performance between the 2 versions.
v1.7.3
https://logs.px4.io/plot_app?log=6857323d-fe61-406d-b00e-5905a90ce591
v1.8.0
https://logs.px4.io/plot_app?log=601fe7fa-b820-46bd-8332-cc2d4ac800f4

*The versions referenced are the stable versions.

[priseborough]

Unfortunately the change to the cmake system between the ecl versions used for versions 1.7.3 and 1.8.3 meant that I was unable to build 1.7.3 using the ecl library from 1.8.0.

[AuterionWrikeBot]

➤ Wrike Bot commented:

@paul Riseborough All subtasks have been completed or deferred, you might want to complete this task as well.

[priseborough]

I was able to compare the v1.7.3 and v1.8.0 logs and check the EKF baro lat innovations and the consistency of the output vertical velocity and vertical position derivative. Both metrics indicated that the ekf in the 1.8.3 log was performing better than in the 1.7.3 log in terms of bias errors, but had more noise on the baro innovations.

v1.8.0 baro innovations

v1.7.3 baro innovations

v1.8.0 vertical velocity and vertical position derivative

v1.7.3 vertical velocity and vertical position derivative

[priseborough]

@jfcisneros Plotting the baro alt and distance sensor data overlayed shows a significant inconsistency. Can you please confirm your sensor calibration. If the baro has this much scale factor error, then it is pointless doing 1.7.3 vs 1.8.0 comparisons.

v1.7.3

v1.8.0

[tops4u]

Could the Barometer be exposed to Downwash/Turbulence? I'm interested in this Issue since I have also seen rather large Oscillations of my F-450 Quad with a Pixracer, but now when I think of it, it is probably exposed to Downwash Effects (it does not have LIDAR/US Sensors)

[priseborough]

That's possible. A hand held height change test without motors running will confirm if it is a sensor or aerodynamic issue.

[priseborough]

I am closing this issue as there is a clear problem with baro data in flight. Once this has been resolved, the issue can be reopened if the problem with altitude hold with 1.8.0 continues.

[jfcisneros]

Sorry for the delay. I just did the recommended testing without the motors running.
Both of the logs were recorded with v1.8.0
Hand Held Test
https://review.px4.io/plot_app?log=4c726610-db8d-472b-924f-2359e13b0f54

Altitude Hold
https://review.px4.io/plot_app?log=074083e5-d457-47fb-8f96-f16c36ace664

There seems to be a great effect in the baro scale factor while flying.
For our vehicle we have selected Range sensor as EKF2_HGT_MODE, and Range aid enabled for EKF2_RNG_AID. Wouldnt this set the EKF2 to use Range Finder height rather than height from the barometer?