Bodies with _only_ visual geometry -- are they supposed to participate in the simulation?

[rao94583]

Intro

We're using MuJoCo for some robotics research (but then again, who isn't?)

My setup

MuJoCo on MacOS and Ubuntu.

My question

In the ur5e.xml file in mujoco_menagerie, there are a body element (base) that contains only an inertial element and a geom element of class visual. I think this means that only the inertial element interacts with the physics engine, and the geometry is only used for visualization. And this further implies that a body must contain at least an inertial element, or a single non-visual geom element.

Am I thinking about this right?

Minimal model and/or code that explain my question

https://github.com/google-deepmind/mujoco_menagerie/blob/main/universal_robots_ur5e/ur5e.xml

Confirmations

• I searched the latest documentation thoroughly before posting.
• I searched previous Issues and Discussions, I am certain this has not been raised before.

[yuvaltassa]

Not exactly.

A visual (non-colliding) geom could be used for inertia inference, and in that case would affect the physics. In the case you described, where inertia is explicit, this geom will indeed not affect the physics.

To clarify further, geoms can affect the physics in 3 unrelated ways:

1. Collisions
2. Inertia inference.
3. Ray casting targets (e.g. rangefinder sensors)

As you see, the notion of "visual geom" which comes from URDF, is not really a helpful one and should be avoided. For example a geom that URDF would call visual will affect sensor values if a ray is cast upon it. Does that count as "affecting the physics"? 🤷🏼‍♂️