Isaac Sim–generated datasets meet an MLS‑MPM/CPIC cutting engine for physically grounded learning.
•
Mesh → SDF Pipeline
Convert meshes to Signed Distance Fields (SDF), with automatic voxel size adjustment, padding, and sign correction. Generate blade_mask based on blade axis and fractional offset.
Module: mesh_sdf
•
Colliders (Knife/Board)
Provide SDF sampling, surface normals, masks, and velocity fields. Support knife y-axis oscillation and z-offset adjustments. Includes classifier (mesh vs. board).
Module: colliders
•
Elasticity/Plasticity
Decomposition into deviatoric and hydrostatic components, Frobenius norm evaluation, and J2 radial return mapping (with isotropic hardening and viscoplastic additive options).
Physics module
•
MPM Kernels
Particle-to-grid (p2g) and grid-to-particle (g2p) transfers, corotated elasticity, CFL condition enforcement, velocity capping, J-determinant clamping, and knife/board contact responses (friction and restitution).
Module: sim
•
Rendering/Overlay
Update end-effector position, force indicators, and cutting band coloration. Provide viewer and GUI utilities.
Module: renderer
•
CLI / Output
Support preview and headless modes, logging/export utilities, camera presets, and toggles for visualizing material damage.
Module: cli/output
SDF Demos:
Demo video:
Export JSON:
{
"timestamp": 12.433,
"episode_id": "epi_000123",
"camera": {"id": "cam_front", "K": [...], "T_world_cam": [...], "resolution": [1280, 720]},
"rgb_path": "frames/epi_000123/cam_front/000372_rgb.png",
"depth_path": "frames/epi_000123/cam_front/000372_depth.exr",
"ee": {"pos_m": [x,y,z], "quat_xyzw": [x,y,z,w], "normal_world": [nx,ny,nz]},
"knife": {"y_anim": 0.184, "z_offset": -0.036},
"force_world_N": {"x": 1.23, "y": -4.56, "z": 0.78, "norm": 4.73},
"material": {"id": "banana_v3", "density": 400.0, "youngs_E": 3.0e5, "poisson_nu": 0.45},
"rand_seed": 1874329
}
Python
복사
