CollisionChecker accounts for non-floating bodies with floating joints

[SeanCurtis-TRI]

Fixes #21215

---

This change is 

[calderpg-tri]

Reviewable status: 1 unresolved discussion, needs platform reviewer assigned, needs at least two assigned reviewers, missing label for release notes (waiting on @SeanCurtis-TRI)

---

planning/linear_distance_and_interpolation_provider.cc line 27 at r1 (raw file):

    const RigidBody<double>& body = plant.get_body(body_index);
    // We only grab the quaternions for floating bodies .
    if (body.is_floating() && body.has_quaternion_dofs()) {

This doesn't make sense to me. If you have a body with quaternion DoF that is not floating, this now guarantees that the interpolation and distance will be wrong.

[jwnimmer-tri]

Reviewable status: 1 unresolved discussion, needs platform reviewer assigned, needs at least two assigned reviewers, missing label for release notes (waiting on @SeanCurtis-TRI)

---

planning/linear_distance_and_interpolation_provider.cc line 27 at r1 (raw file):

Previously, calderpg-tri wrote…

This doesn't make sense to me. If you have a body with quaternion DoF that is not floating, this now guarantees that the interpolation and distance will be wrong.

Agreed

[SeanCurtis-TRI]

Reviewable status: 1 unresolved discussion, needs platform reviewer assigned, needs at least two assigned reviewers, missing label for release notes (waiting on @SeanCurtis-TRI)

---

planning/linear_distance_and_interpolation_provider.cc line 27 at r1 (raw file):

Previously, jwnimmer-tri (Jeremy Nimmer) wrote…

Agreed

Oops. I pushed it the wrong direction. The goal wasn't to be more exclusive, it was to be more inclusive. My bad.

And the RigidBody api doesn't have an API for positions_start().

But, in the meantime, you can steal the test and call it good.

[SeanCurtis-TRI]

This has served its purpose. I'm closing it.