IMU-bias frame

[thomash...@gmail.com]

I was debugging some performance regressions and stumbled upon a possible correlation between the yaw and the imu-bias (as it is optimized using the `CombinedImuFactor`)

See attached screen that shows the yaw on the bottom plot and the x-y bias on the accelerometer. I would have expected the IMU-bias to be kind of independent of the yaw-orientation; however this seems to either suggest it is not (or something else is wrong). Can someone confirm the bias under normale circumstances should be in the imu-sensor frame (and hence independent of the navigation frame)?

[Brice Rebsamen]

hello

Yes this is correct, the IMU biases are in the imu-sensor frame.

Depending on your setup, it is possible that what you are observing is
that the IMU biases are becoming more observable as the robot is
moving around and turning: the acc biases are not observable directly
unless there is some motion that can help disambiguate the biases from
real accelerations, and accelerations due to gravity bleeding onto xy
axes due to roll/pitch. Perhaps try with a batch solve first to see
whether you have the same effect, and if yes then perhaps you have
something wrong in how you setup your graph...

Brice

> --
> You received this message because you are subscribed to the Google Groups "gtsam users" group.
> To unsubscribe from this group and stop receiving emails from it, send an email to gtsam-users...@googlegroups.com.
> To view this discussion visit https://groups.google.com/d/msgid/gtsam-users/8e44c83d-b36a-4893-b59c-bdce0f3b75een%40googlegroups.com.