BNO085 unstable roll value

I have Upgraded my IMU from CMPS14 to BNO085 and now it seams that the roll value is not absolute anymore.
It varies over time about 3-4 degrees in every direction. With the CMPS which uses the magnetometer it was always rock solid!

Has anyone experienced something similar with the BNO085 used with PANDA over USB?
Can it have something to do with “Game Rotation Vector Mode”?

Thank you!

When the tractor stands still on a sidehill the roll value will kind of be “drawn” towards zero. I tried it for fifteen minutes and it was sowly going from 3,5 down to 1,2 without a single movement of the IMU.
Is this behavior typical for an IMU that isn’t corrected via magnetometer?

As I’ve got my antenna mounted at 3m of height it significantly influences pass-to-pass accuarcy.
To eliminate that I have to alter my implement offset setting several times in one job to keep the error under 10cm…

It’s a bit frustating because with CMPS14 roll worked great!

I am having similar issues as a temporary fix you can set your antenna height to 0 to take the roll out the equation.

Are these BNO from Adafuit/Mouser/Digikey or from other supplier?

All my Adafuit BNO085 are super stable.

My BNO is from Adafruit. I noticed that the value is fairly stable when I zeroed the BNO at 0,00 degrees. When it is set at a different angle the error seams to add up and it gets worse. Might be a Software issue from AOG?! I have the roll inverted, too.

The roll as well dances around when I drive around the corner as if the centrifugal forces influence the roll in “Game rotation Vector Mode”

I have also moved from cmps to bno and on the concrete where I zero roll the bno it’s changing from 0 to 3-4 degrees when I slightly move rhe tractor on the concrete.
Should I change something on the ino files when I switch from cmps to bno?
Also can I add my old cmps on the panda board without unplugging the bno?

Check if axis is correct.

My bno from ali seams stable.

Why not use multiple IMU 3 to 4, discard most deviating value and average other 3. If magnetometer (rotation vector mode) auto calibration is at consistent intervals then they can be delayed turn on so they dont recalibrate at same time. And use magnetic field for more stable roll, heading value?

Or is there a reason this wouldn’t work?