[BUG] [imu.py] gravity in world frame was subtracted from lin acc in body frame

[pschoc]

Description

Related Issue

Resolves Genesis-Embodied-AI/Genesis#

Motivation and Context

How Has This Been / Can This Be Tested?

Screenshots (if appropriate):

Checklist:

• I read the CONTRIBUTING document.
• I followed the Submitting Code Changes section of CONTRIBUTING document.
• I tagged the title correctly (including BUG FIX/FEATURE/MISC/BREAKING)
• I updated the documentation accordingly or no change is needed.
• I tested my changes and added instructions on how to test it for reviewers.

• I have added tests to cover my changes.
• All new and existing tests passed.

[pschoc]

also I have trouble understanding the skew:

how is the "skew" meant to be used or what is it expressing? From what I see in imu.py/_get_skew_to_alignment_matrix() it generates a rotation matrix of the form:
[ 1 v1 v2 ]
[ v0 1 v2 ]
[ v0 v1 1 ]
given a skew vector [v0, v1, v2]. That makes no sense to me, I was expecting sth like eye(3) + hat([v0,v1,v2]) =
[ 1 -v2 v1 ]
[ v2 1 -v0 ]
[ -v1 v0 1 ]
or handing over the vector of angular misalignments which then get mapped from euler angles to so3

[duburcqa]

@Milotrince Why the unit test is passing ?

[Milotrince]

@pschoc Thanks for the catch.

@duburcqa I will update the unit test to cover this, i think the current test is too simple (local orientation = global orientation) so it didn't catch it.

also I have trouble understanding the skew:

The axis skew is supposed to represent sensor axis misalignment / cross-axis coupling.

matlab also uses this definition, for reference: https://www.mathworks.com/help/nav/ref/imu.html

[Milotrince]

maybe we could rename axes_skew to axes_misalignment or axes_coupling to avoid confusion ?