Primvaria

Primvaria · Granulaire

Limits

Everything Granulaire does not do or has not verified, stated plainly.

Limits

  • Friction is per pair of surfaces. What a robot foot feels also depends on the foot's own material and the engine's combine mode.
  • No capillary cohesion in simulation. Wet-grain cohesion is stored in granulaire:capillaryCohesion, but OpenUSD has no standard schema for it and no rigid-body engine reads it: in simulation, wet grains differ only by friction, restitution and mass.
  • No rolling resistance. Neither UsdPhysics nor PhysX rigid bodies model it, which is why piles of spheres flow far too easily. Here the irregular shapes play that role: an angular grain cannot roll. Rounded grains (river pebbles, beads, soybeans, pellets) do roll: add the engine's own rolling resistance when it has one (the MuJoCo test does).
  • A pack is a sample of the grading curve. Sizes are drawn from a normal law in phi units (phi_mean, phi_std of the preset), and the tests check 2000 draws per preset against it. A pack of 100 shapes is a small sample: its grading entry in manifest.json gives the sample's mean and spread next to the target's, with the Kolmogorov–Smirnov statistic and p-value. For a closer match, generate more shapes.
  • Calibrate for quantitative work. For training with domain randomisation, these values are a sensible centre to randomise around; for a quantitative study, calibrate them against the real material, for example with its angle of repose.
  • Angle of repose not verified. The pile test proves that grains settle and stay put, not that a heap takes the real material's slope. Two CPU tests were tried in MuJoCo 4.0.1 and neither can be trusted. A heap poured from a lifted cylinder (200 crushed-gravel grains, 15 minutes) came to rest at about 21°, well below what crushed stone does. A tilting bed (54 to 96 free grains on a layer of fixed ones) started to slide at 31° for crushed gravel and 33–35° for glass beads, the reverse of the real materials' order. A few hundred grains is too few for a slope to mean anything, and more does not fit in a CPU check. No angle of repose is published for any preset until a test reproduces known materials.

Known limits

  • Real units. See Physics limits by grain size above.
  • No grain-by-grain physics at large scale. CPU physics targets a few hundred grains; instancing is for rendering surfaces.
  • Angularity is a proxy, not Wadell roundness, and is only comparable between grains of the same mesh resolution.
  • Collision uses a convex hull of 64 points for rigid grains, within 2.4 % of the grain's size of its full hull. Pits, creases and dents are visible but not part of the collision shape; hollow shells (whole seashells) are not generated. In the OmniPhysics layer, deformable grains collide with their tetrahedral mesh.
  • Deformables work only with NVIDIA OmniPhysics, untested. Until OpenUSD standardises deformable bodies, packing peanuts and cherry tomatoes are simulated as deformables only in Omniverse / Isaac Sim (GPU), through the OmniPhysics layer over the file; elsewhere they are rigid bodies. They have not been run in Isaac Sim yet.
  • Mesh resolution. Marching cubes runs on a 48-cell grid along the longest axis; the sharpest edges can show slight stair-stepping.