Research

What we work on

Automating the construction process end to end, in six stages.

The objective is a chain that runs from the survey of what exists to a finished building that keeps reporting its own data, without a manual transfer between any two stages. Today each stage needs someone to redraw, re-enter or re-measure what the stage before it produced.

Open questions

Operational state

A model that reports what the building is doing now, when it last observed it, and how confidence decays between observations.

Federated twins

Several twins on one project, with different owners and update rates, made to agree where it matters without being merged into one model.

Pre-semantic scene proxy

A representation between raw capture and classified model: measurable, navigable, enough to drive a machine, produced before classification.

Long-horizon environmental twins

Twins meant to stay current for years rather than for the length of a contract.

Selection

What gets into this list

We follow one chain: survey what exists, turn it into a model that stays faithful, drive a machine from that model, and measure what came out. A paper enters the list when it reports a working step of that chain — with a site, a machine or a dataset behind it — and when it appeared in a venue we read regularly: Automation in Construction, Advanced Engineering Informatics, Developments in the Built Environment, Construction Robotics, the ISPRS journals and archives, and arXiv preprints in robotics and computer vision.

Reviews are in only when they map a step we work on. Position papers without a result are out. Our own papers are in the list and marked as such.

The full collection, with filters by year, stage and author →