SDF grinding in Newton

A small example of SDF-SDF contact and geometric material removal. It does not run a dynamics solver.

Overview

• The workpiece is a static ellipsoid stored as a texture SDF.
• The grinder is a kinematic cylinder. Both shapes have hydroelastic contact enabled.
• Each frame calls the hydroelastic collision pipeline directly.
• ViewerGL renders the contact surface and plots its integrated pressure.
• The grinder volume is subtracted from the workpiece with an SDF CSG difference.
• The zero isosurface is rebuilt with Marching Cubes and logged as a dynamic ViewerGL mesh.

Video

ViewerGL shows the moving tool, updated workpiece, pressure-colored contact field, and integrated normal-load plot.

Set up SDF-SDF contact

• Build the workpiece SDF explicitly so its samples can be edited later.
• Set is_hydroelastic=True on both shape configurations.
• Request the contact surface for ViewerGL.
• No solver is required to call the collision pipeline.

workpiece_sdf = workpiece_mesh.build_sdf(
    max_resolution=256,
    narrow_band_range=(-0.08, 0.08),
    texture_format="float32",
    paired_samples=False,
)

shape_cfg = newton.ModelBuilder.ShapeConfig(
    is_hydroelastic=True,
    kh=1.0e8,
    gap=0.005,
    margin=0.0,
)

pipeline = newton.CollisionPipeline(
    model,
    sdf_hydroelastic_config=newton.geometry.HydroelasticSDF.Config(
        output_contact_surface=True,
    ),
)
contacts = pipeline.contacts()
pipeline.collide(state, contacts)

viewer.show_hydro_contact_surface = True
viewer.log_hydro_contact_surface(
    pipeline.hydroelastic_sdf.get_contact_surface()
)

Remove material

• For an SDF difference, use max(workpiece_distance, -tool_distance).
• Apply the operation to the coarse samples and every allocated fine subgrid sample.
• Copy the modified arrays back into the Texture3D objects.
• Extract and log a new surface after each edit.

point_in_tool = wp.transform_point(
    wp.transform_inverse(grinder_xform), point
)
tool_distance = _sdf_cylinder_z(
    point_in_tool, grinder_radius, grinder_half_width
)
sdf_values[z, y, x] = wp.max(sdf_values[z, y, x], -tool_distance)

texture_data.coarse_texture.copy_from(coarse_values)
texture_data.subgrid_texture.copy_from(subgrid_values)

surface = workpiece_sdf.extract_isomesh(device=model.device)
viewer.log_mesh(
    "/workpiece_sdf", points, indices, dynamic=True
)

Notes

• This is geometric removal only. Contact forces do not drive the grinder.
• The built-in contact-surface color is based on penetration depth. With the same linear stiffness on both shapes, it is proportional to pressure.
• The plotted value is integrated pressure, or normal load: sum(max(-distance * stiffness, 0)).
• A physical wear model could scale removal by pressure, slip speed, and time step.
• Higher SDF resolution increases both the number of edited samples and the cost of rebuilding the Marching Cubes mesh.
• The example requires a CUDA-capable device.

Run

cd <newton-checkout>
uv sync --extra examples
uv run python <path-to-download>/example_sdf_grinding.py

Full source

Loading example_sdf_grinding.py…