BEYOND PROXIMITY
SEMANTIC 3D SCENE GRAPHS
2026 — /002
A hierarchical semantic graph that prunes scene search before grounding natural-language queries in captured evidence — retrieval that understands structure, not just distance.
RESEARCH
PAPER SUBMITTED TO IEEE — COMING SOON
THE QUESTION
A captured space holds far more than any system can reasonably read at once. Ask it something in plain language and it will search everything — every corner, every viewpoint — to answer what a person would have narrowed down in a second.
Beyond Proximity starts from how people actually move through a place: a building holds areas, areas hold things, and every thing was seen from somewhere. The work asks whether that structure alone can carry a query most of the way to its answer.
GROUNDED IN WHAT WAS SEEN
Answers stay tied to the evidence they came from — the specific views a claim rests on, not a confident sentence with nothing behind it. The trade-offs are measured and reported honestly, including where the approach still costs something.
WHERE IT STANDS
Evaluated on indoor scenes captured by the team, alongside pilots on established public datasets. The method, the measurements, and the reproduction material stay held back until publication.
Preprint with Vizan, Shankin, Nuruzov, and Medvedev, 2026. The paper is submitted to IEEE; the public write-up follows it.