Sensing in housing: what a dwelling should be able to answer

We were asked what we can speak to on sensing in housing: detecting occupancy and condition through the fabric of a building, opening and closing apertures automatically, and the security and crime-prevention side of putting sensors into places people live. The answer runs through one structure, so it is set out here in full.

The ground we stand on. We have worked with IoT in buildings for over a decade, held responsibility for digital twin strategy, and developed self-learning, platform-independent heating control that has to account for the contextual behaviour of whichever building it is placed in. That last constraint is the relevant one. A dwelling is a thermal, acoustic and electromagnetic object before it is a data source, and a control loop that ignores the fabric it sits in will be re-tuned by hand for the rest of its life.

Occupancy and condition read through the fabric. Presence, movement, breathing and heart rate leave signatures in the ordinary physical channels a dwelling already carries: radio propagation between fixed nodes, thermal gradients, acoustic response, differential pressure across apertures, and the load signature on the electrical circuit. Reading them through the fabric removes the camera from rooms where a camera is unwelcome, which is most of a home. The same signals drive apertures: windows, vents, dampers and shading that open and close on measured conditions rather than on a schedule.

Where the intelligence sits determines what the system is. Our position is that residential sensing converges on four properties. It is self-learning, so the installer is not the person who tunes it. It is platform-independent, so a component can be replaced without replacing the system. It is energy-harvesting wherever the physics allows, so the maintenance burden of a sensor population stays bounded. And the inference and the control loop run locally, so the dwelling continues to work when the connection does not.

Standards. Matter is the interoperability layer with the broadest commitment behind it, and it is the one we build against for device-level interworking in the home. Alongside it we watch mioty for its telegram-splitting robustness in dense radio environments, and LoRa used in mesh topologies for coverage through structure without a single gateway to lose. Our read, and it is a read rather than a published finding: mioty's adoption curve has been slower than its technical case, and the residential market will settle on a small number of decentralised, peer-to-peer capable stacks rather than on one.

The reasons are regulatory, operational and adversarial, in that order. The EU AI Act places obligations on systems that make consequential inferences about people, and inference that never leaves the building is a materially different compliance object from inference shipped to a third party. Resilience: a dwelling that cannot open a vent during an outage has a safety property depending on someone else's uptime. Attack surface: every device that reaches outward is a route inward, and a local-first topology shrinks that surface by construction.

Across the lifecycle, the model outlives the software. BIM, GIS and CIM together carry the geometry, the context and the civic layer, and they survive the tools that produced them. Held that way, a building can answer questions about its past, its present and its projected future to a person, to another system, or to an AI agent, without any of them needing access to the raw record. We have built twins of buildings, rooms, energy systems, plants, cities, telecom networks and of people, and the structure repeats at every scale.

Security and crime prevention. Cameras and shutters are the visible half. The other half is design, occupancy pattern and behaviour: sightlines, lighting, the appearance of presence, and the detection of the unusual against a learned baseline of the ordinary. Multi-agent systems are what make the second half operable, because a rule set that has to be written by hand cannot keep up with how a household actually lives. The governing constraint is that the underlying data stays with the household. What leaves is an anonymised insight, and the difference between those two things is architectural.

What follows for a homeowner, and for anyone selling into the home. The properties that decide the outcome are resilience, data ownership, an explicit data strategy, security against the cryptographic horizon including post-quantum migration, personalisation, decentralisation and anonymisation. They are the same list from either side of the transaction, which is unusual and is the reason the segment is moving.

The method behind this is published openly with a DOI, and the ordering it imposes is the argument in one line: goal, benefit and impact first, then decisions, then insights, then information. Data last.