Red Wall Kitchen — Parametric CAD from RGB Video

A real kitchen captured via ARKit video is reconstructed as a fully parametric CAD model using Build123d (OpenCASCADE). Instead of noisy mesh extraction from Gaussian splats, every surface is a clean watertight solid — measured from 3DOD annotation boxes and grounded to LiDAR dimensions.

Modular generators produce each kitchen zone independently:

West wall: diamond mosaic backsplash, base cabinets, granite counter, cooktop, double oven, glass uppers

North wall: window, 1.5 sink with faucet, raised-panel base, arched-glass upper cabinets

Entrance: thick wall with articulated 6-panel door and hinge

Dining: table, 4 chairs, wall shelves, radiator

Each generator outputs a STEP file with a materials.json sidecar (per-part friction and density). The assembly composes into a single labelled STEP, then exports to GLB for visualization and USD/MJCF for physics simulation.

This is CAD Kitchen v3 — the model-driven complement to the data-driven splat pipeline (Lab Study 2, Kol A). Clean geometry means reliable colliders for robot contact tasks.