About the quaternion convention

[simon-bachhuber]

Suppose one starts with the docstring from base.Geometry

    transform: transform from this geometry's coordinate space to its parent
      link, or to world space in the case of unparented geometry

And now we use this transform to rotate a vector

r_parent = math.rotate(r_geom, transform.rot)

What is the interpretation here? Is

• r_parent the old vector r_geom but in basis of parent
• or is r_parent a new rotated vector in basis of geom frame

Note: These two interpretations are related by inverting transform.rot

[btaba]

geometry.transform is the position/orientation of the geom in the parent frame, that docstring is a bit confusing
Here's an example of usage:

brax/brax/geometry/contact.py

Line 499 in ff3ff64

tx_i = x.take(geom_i.link_idx).vmap().do(geom_i.transform)

so in your case, r_parent is r_geom rotated by transform.rot in the parent frame