# Environment textures and geometry

The unmodified color PNGs are from ambientCG, created by Lennart Demes and released
under [CC0 1.0 Universal](https://docs.ambientcg.com/license/). They may be bundled
with the simulator and used in rendered training data. Attribution is preserved
here for provenance although CC0 does not require it.

| Local file | Source asset and archive member |
|---|---|
| `oak_color.png` | [Wood049](https://ambientcg.com/view?id=Wood049), `Wood049_1K-PNG_Color.png` |
| `pale_wood_color.png` | [Wood050](https://ambientcg.com/view?id=Wood050), `Wood050_1K-PNG_Color.png` |
| `dark_wood_color.png` | [Wood051](https://ambientcg.com/view?id=Wood051), `Wood051_1K-PNG_Color.png` |

All three are procedural wood materials with nominal 80 x 80 cm source coverage.
Only their color maps are used; no claims of normal mapping, calibrated roughness,
or physically based rendering are made for the classic MuJoCo renderer.
`manifest.json` records download URLs and SHA-256 hashes of the vendored files.

The room, worktable, robot enclosure, camera mast, cable routes, board inlays and
tray details are authored in `sim/scene_environment.py`. The robot/chess meshes
have separate provenance in their respective asset directories. Cosmetic hardware
uses visual geom group 1; hidden contact proxies use group 3. The finite tabletop
has a box collision envelope with its top at z=0. Bevelled visual edges differ from
that envelope by at most 3 mm. The enclosure retains its original box envelope.

The scene is a plausible tabletop setup at the existing arm/board scale. Its
lighting, materials, cable routes and furnishings are not measurements of a real
deployment site. Camera pose changes are documented in `sim/README.md`.