Nested models which don't contain the canonical link will not have accurate poses

[brawner]

In a nested model, poses are relative to the canonical link, which may be in another model. This prevents a nested model that doesn't contain the canonical link from having a correct pose and it means that the link inside that model has a pose relative to the canonical link in another model.

In this example, one sphere starts off 1 meter above an inclined plane. The canonical link is model_00::model_10::link_00. When simulating, the pose of model_00::model_11::link_01 changes and is relative to model_00::model_10::link_00, while the pose of model_00::model_11 remains its originally specified pose.

I would expect the pose model_11 to update and the pose of link_01 to remain 0 in this case.

PoseToCanonicalLink.mp4

[brawner]

@EricCousineau-TRI per today's meeting

[brawner]

This issue follows up #542

[EricCousineau-TRI]

I think I understand now - so the physics are as expected (both spheres's move in sensible ways), but the GUI's indication of the non-canonical-link-containing-model (and it's link) are unexpected.

So the solution would be to tell the GUI / Physics engine what to express those links relative to?

And the other question - is the time-varying position of model_00 what you expect?

[brawner]

Yes to both questions. model_00 definitely updates as expected, since it follows the canonical link. model_10 is also correct because the reference frame of link_00, model_10 and model_00 are all identical. The issue is with models and links that don't contain this canonical link.

[azeey]

I believe this has been fixed since #685. I've reproduce the example in the issue in case anyone else wants to test it out:

pose_to_canonical_link.sdf.txt