Build an indoor routing graph from Matterport vector data: doorways plus 'invisible' edges
From the Model API rooms{boundary{edges{type openings vertices}}} data you can build a walkable graph automatically. Edge type is 'wall' or 'invisible'; invisible edges are open gaps between rooms, and rooms like a kitchen or coworking area can have zero doorways yet be connected through them, so ignoring invisible edges leaves rooms unreachable. Opening types include 'doorway' and also 'window', so filter windows out. Compute each opening's world position as v0 + relativeCenter*(v1-v0), merge openings from different rooms within about 0.6 m into one door (and invisible-edge midpoints within about 0.8 m into one passage), polygonize each room's edges with shapely, and join a room's doors pairwise only if the straight segment stays inside the polygon buffered by about 0.4 m. Openings seen from only one room are doors to unscanned space: skip them. Coordinates are relative to each scan's own origin, so floors and stitched areas still need manual alignment, and the thumbnail (model{image{url}}) tells you what the space really is.