Unable to correctly set inertia when randomizing COM in Isaac Lab

[YeJiAce]

Question

Hello, I am doing domain randomization on a simple rigid box in Isaac Lab.
My workflow is:

1. Randomly shift the object’s COM.
2. Recompute the inertia tensor using the parallel-axis theorem.
3. The resulting inertia matrix is non-diagonal.
4. I apply it with: asset.root_physx_view.set_inertias
5. However, I found that all off-diagonal terms are silently discarded, and only the diagonal values remain.
6. According to the PhysX documentation, inertias are defined in the rigid body’s COM frame, and in this frame, the inertia of a box should indeed be diagonal.

My question is:

Since the inertia tensor is defined in the COM frame, if I shift the COM, does that mean I don’t need to change the inertia tensor? And is this physically consistent with real-world object dynamics in simulation?

Thanks in advance.

• Isaac Lab Version: [2.2.0]
• Isaac Sim Version: [5.0.0]

[RandomOakForest]

Thank you for posting this. The short answer, if I understand your question correctly, is no: if you change the COM location relative to a fixed body frame, the inertia tensor for that body frame is no longer the same and must be updated via the parallel‑axis theorem.¹² I'll move this post to our Discussions for follow up. Here is a summary to consider.

What stays the same

The inertia tensor about the COM, expressed in a frame whose origin is at the COM, does not change just because you describe the COM as being at a different place in some other frame.³¹
So if you keep a frame that always sits at the COM, and you only “move the COM” by redefining where that frame is relative to a world or actor frame, then the inertia tensor in that COM‑attached frame stays the same.¹³

What changes when you “shift COM”

From the perspective of a fixed body frame (for example, the actor frame used to define collision geometry), moving the COM means you are now interested in the inertia about a different point.²¹
The parallel‑axis theorem states that the inertia about this new point equals the inertia about the COM plus a mass‑dependent term, so the inertia tensor in that fixed frame must be updated accordingly.⁴²

How this maps to PhysX / Isaac Lab

PhysX stores inertia in the mass frame: origin at COM and axes aligned with principal moments, where the tensor is diagonal.
When you “shift the COM” in PhysX (e.g., using PxMassProperties::translate), PhysX simultaneously updates the inertia tensor so that the stored inertia remains the correct one for the new COM frame; the helper API does the parallel‑axis bookkeeping for you.

So, in words matching your question:

• If “shift the COM” means “I re‑choose the reference point but still want inertia in that new COM frame” → the inertia tensor values can change when you do the physically correct translation, and PhysX expects those updated values.
• If “shift the COM” means “I only change where this COM‑attached frame sits relative to world, but I keep the frame origin at the COM” → the inertia tensor in that COM frame does not change.

For Isaac Lab, when working with set_inertias (mass‑frame diagonal) and COM pose separately, you typically keep inertia diagonal in the COM frame and modify only the COM pose for your randomization, which is physically consistent with how PhysX models rigid bodies.

Footnotes

1. https://phys.libretexts.org/Bookshelves/Classical_Mechanics/Variational_Principles_in_Classical_Mechanics_(Cline)/13:_Rigid-body_Rotation/13.08:_Parallel-Axis_Theorem ↩ ↩² ↩³ ↩⁴

2. https://pressbooks.library.upei.ca/statics/chapter/inertia-intro-parallel-axis-theorem/ ↩ ↩² ↩³

3. https://galileoandeinstein.phys.virginia.edu/7010/CM_23_Rigid_Body_Motion.html ↩ ↩²

4. https://engineeringstatics.org/parallel-axis-theorem-section.html ↩