Make the tiled ground plane the default

[maxkra15]

Description

Bundle a warm-white ground plane with NVIDIA-green metric grid lines and make it the default for GroundPlaneCfg and plane terrains.

This change:

• replaces the remote Nucleus default with a repository-owned USD and textures;
• preserves the authored floor appearance unless callers explicitly provide a color tint;
• makes authored face-varying primvars:st the sole texture-mapping authority for Kit, Newton RTX, and Newton GL;
• configures OmniPBR to consume UV set 0 (st) with projection disabled and identity texture scale;
• recomputes the bundled asset's UVs when GroundPlaneCfg.size changes so its 5 m texture tile and 1 m grid stay metric;
• sizes a plane terrain's visual mesh to cover its environment grid, with a 100 m minimum, while retaining an infinite collision plane;
• keeps the historical dark material default for generated terrains; and
• documents and tests the asset contract.

The bounded visual mesh matters for standalone visualizers: the previous 2,000 km mesh exhausted useful UV precision near the origin and could sample one green texel across the entire view. Collision behavior is unchanged because the USD Plane collider is infinite.

This is the asset-level solution that superseded #7352. It does not require a Newton-specific OmniPBR compatibility patch or Nucleus upload. Renderer tone mapping and light models can still produce different final brightness.

Renderer comparison

All clips use the bundled default floor, one environment, 300 frames at 60 FPS (5 seconds), and a fixed camera per task. Each renderer panel is captured at 1280 × 720. Kit uses PhysX; Newton RTX and Newton GL use MJWarp.

To keep the warm-white floor below clipping and make the grid legible, these presentation captures use fixed neutral calibration: Kit uses each environment's neutral scene light at 40% of its normal intensity, Newton RTX uses a neutral dome at intensity 400, and Newton GL uses the established neutral ambient profile at exposure 0.40. This calibration affects the media only and is not part of the runtime ground-plane change.

ANYmal-D walking policy

Full-resolution ANYmal-D MP4

Cartpole policy

Full-resolution Cartpole MP4 · Both environments in one MP4

Type of change

• New feature (non-breaking change which adds functionality)
• Documentation update

Release backport

• Backport this pull request to the active release branch after it merges into develop

Testing

• uv run --frozen --no-sync isaaclab -f
• uv run --frozen --no-sync pytest source/isaaclab/test/sim/test_default_ground_plane.py -q (3 passed)
• uv run --frozen --no-sync pytest source/isaaclab/test/sim/test_spawn_from_files.py::test_spawn_ground_plane -q (1 passed)
• uv run --frozen --no-sync pytest source/isaaclab/test/terrains/test_terrain_importer.py::test_plane -q (4 passed)
• Manual Cartpole and ANYmal-D policy playback with Kit/PhysX, Newton RTX/MJWarp, and Newton GL/MJWarp

Checklist

• I have read and understood the contribution guidelines
• I have run the repository formatting and pre-commit checks with uv run --frozen --no-sync isaaclab -f
• I have made corresponding changes to the documentation
• My changes generate no new warnings
• I have added tests that prove my fix is effective or that my feature works
• I have added a changelog fragment under source/<pkg>/changelog.d/ for every touched package
• My name already exists in CONTRIBUTORS.md

[ooctipus]

Thanks—this asset-level approach looks good, and the authored metric UVs are the important part. Could we make primvars:st the single source of texture mapping for every renderer, rather than maintaining both OmniPBR projection and equivalent authored UVs?

Newton #4007 is now merged. Newton supports the standard OpenUSD path:

UsdPrimvarReader_float2("st") -> [optional UsdTransform2d] -> UsdUVTexture -> UsdPreviewSurface

That change intentionally does not read OmniPBR mapping inputs (project_uvw, world_or_object, texture_scale, and related attributes). Since this plane already bakes the 5 m repeat into face-varying primvars:st, it does not need a UsdTransform2d either.

My suggestion:

• Keep the metric primvars:st authored and resized by Isaac Lab as the sole mapping authority.
• Author a universal UsdPreviewSurface / UsdUVTexture output with explicit repeat wrapping for Newton.
• If an MDL/OmniPBR output is retained for Kit appearance, have it consume UV set 0 (st) with projection disabled and identity scale, rather than owning a second projected mapping.

This should give Kit, Newton GL, and Newton RTX the same tiling while preventing the two mappings from drifting. Is there a Kit-specific reason this asset still needs object projection instead of consuming the authored st?

[ooctipus]

One clarification on the maintenance boundary: the current dual-authoring approach—OmniPBR object projection plus equivalent authored UVs—is not ideal because the same tiling semantics have two sources of truth and can drift.

Making primvars:st the sole mapping source removes that duplication. However, if both an OmniPBR material output and a UsdPreviewSurface output remain, there are still two material descriptions to maintain even though they share one UV mapping.

The cleanest end state, if Kit provides the required visual fidelity, is therefore a single standard UsdPreviewSurface / UsdUVTexture graph consuming primvars:st across Kit, Newton GL, and Newton RTX.

[ooctipus]

Pushed 261ca78 to this branch rather than leaving the thread open — revert it if you would rather take a different route.

What it does. project_uvw = 0 and texture_scale = (1, 1) on the OmniPBR shader, so it reads UV set 0 instead of projecting. primvars:st becomes the only mapping authority for Kit, Newton GL, and Newton Viewer RTX, and the rescale already in spawn_ground_plane() now keeps the tile metric in all three rather than in two. Asset README and the USD contract test updated with it; the test fails against the projected asset and passes against the new one.

What it does not do, and why I am walking back the second half of my earlier suggestion: adding a universal UsdPreviewSurface output alongside the MDL one would regress Newton. _extract_preview_surface_properties only records a base color when diffuseColor holds a value; with the texture connected it returns color=None, _get_input_value_and_attr then follows the connection to outputs:rgb and gets nothing, and ViewerBase._update_shapes falls back to wp.vec3(0.5, 0.5, 0.5) for a visual-only shape. The GL shader does albedo = ObjectColor; albedo *= tex_color, so the plane would render at half brightness. The MDL path avoids this because diffuse_color_constant = (1, 1, 1) is a plain value, and Newton still resolves the conventional st primvar from it.

So one mapping, still one material description. Worth filing upstream that a UsdPreviewSurface with a textured diffuseColor should default to white; once that lands, dropping OmniPBR for a single standard graph becomes the clean end state I described, subject to the emissive-mask modulation which UsdPreviewSurface cannot express without math nodes.

[matthewtrepte]

hm why is the ground plane rectangular?

[maxkra15]

run-ci