[Bug] Volume deformable extension is almost insensitive to Young’s modulus under attached hanging load

[YangJC112]

Hi NVIDIA team,

I found a possible issue with volume deformables in Isaac Sim 6.0.1.

I tested a simple vertical rigid–FEM–rigid setup:

a volume deformable block is fixed to a rigid body at the top;
a dynamic rigid payload is attached to the bottom;
only gravity is applied;
deformable size: 80 × 80 × 85 mm;
Young’s modulus: 20 / 60 / 150 kPa;
Poisson ratio: 0.35;
density: 180 kg/m³;
payload mass: 0.01–1.0 kg.

The runtime Young’s modulus and payload mass were both read back correctly, and all cases use the same tetrahedral topology.

The unexpected behavior is that the extension changes strongly with payload mass, but is almost independent of Young’s modulus.

In the non-inverted range (0.01 / 0.05 / 0.10 kg), the measured extension-vs-mass slopes are:

20 kPa: 193.2 mm/kg
60 kPa: 190.5 mm/kg
150 kPa: 191.1 mm/kg

These three responses are almost identical.

For example, with a 0.05 kg payload:

20 kPa → 18.90 mm extension
60 kPa → 18.82 mm extension
150 kPa → 18.82 mm extension

I would expect the softer material to stretch significantly more than the stiffer material under the same load. For a simple axial elastic response, the extension should approximately scale with 1/E.

Increasing the payload mass increases the deformation, but does not recover the expected stiffness separation. At higher masses, tetrahedral inversion occurs before a meaningful Young’s-modulus-dependent response appears.

So the main issue seems to be:

mass / load sensitivity is present, but stiffness sensitivity is almost missing.

Is this expected behavior for the current OmniPhysics volume deformable solver, or could this indicate an issue in the deformable–rigid attachment / load transmission path?

I can provide the minimal reproduction script, raw JSON results, and full simulation settings if useful.

I attached two figures showing the mass sweep and the nearly overlapping responses for 20 / 60 / 150 kPa.

[PeterL-NV]

I ran a check that removes the attachments entirely: a block with your dimensions and density, top nodes pinned
kinematically, hanging under its own weight. Closed-form self-weight elongation is δ = ρgL²/(2E), so sag should scale as 1/E:

E	measured sag	analytic
2 kPa	4.282 mm	3.189 mm
5 kPa	1.717 mm	1.276 mm

Going 2 → 5 kPa, measured sag drops by 2.49× against a modulus ratio of 2.50 —
agreement to better than 1%. Raising the load and sweeping your three moduli:

load (ρ)	20 kPa	60 kPa	150 kPa	ratio 20k/150k
180 kg/m³	0.499 mm	0.278 mm	0.258 mm	1.93
900 kg/m³	2.156 mm	0.790 mm	0.352 mm	6.13
1800 kg/m³	4.282 mm	1.430 mm	0.654 mm	6.55

against a theoretical 7.5. So the stiffness response is there.

In my setup the block never fully settles — residual nodal velocities of ~0.05 m/s leave a floor of about 0.26 mm on any extension measurement. In your stiffness range the true elastic extension under a 0.05 kg payload is only 0.33 / 0.11 / 0.04 mm for 20 / 60 / 150 kPa — at or below that floor. That's exactly the top row of my table, where the ratio collapses to 1.93. So a real 1/E response can be completely masked. Worth checking your residual velocity at the moment you sample.

I tried to reproduce your rigid–FEM–rigid arrangement and could not get a clean one: with colliders on the end plates they squashed the block, and without colliders the auto-attachment transmitted no load at all. So I can't yet confirm or rule out the attachment. Could you share:

1. the minimal repro script you offered — in particular how you measure "extension": from the deformable's simulation-mesh nodes, or from the payload rigid body's transform? If it's the latter, it includes any attachment stretch and rigid-body motion, which would explain the magnitude.
2. how the two attachments are created (auto attachment, or explicit attachment prims / tetrahedron filtering);
3. solver iteration counts, substeps and dt;
4. the residual nodal velocity at the time you sample, to confirm it's settled.