BLE trilateration and presence-aware automation

Why BLE rather than the alternatives

The presence-detection options at the household scale are roughly five.

• Motion sensors — cheap and reactive, but they fire on motion, not on presence. Someone sitting still in a room is invisible to them.
• Door sensors — useful for transitions but cannot tell you who is in which room, only that someone went through.
• WiFi-based localisation — works for devices but is too coarse for room-level decisions in a typical house.
• Camera-based detection — accurate but privacy-incompatible with operator-grade household sovereignty.
• BLE trilateration — accurate to room level, privacy-preserving, low power, and works on the devices people already carry.

BLE wins for the same reason most household-scale infrastructure wins: it is the cheapest correct answer that fits the constraints. The signal source is a device the occupant already carries. The receivers are small and cheap. The math is straightforward.

The hardware layout

For a typical home, two to three BLE receivers per floor are sufficient. They are placed at high points, away from large metal surfaces, with line of sight to as much of each floor as possible. We use small ESP-class devices flashed with open-source firmware that publishes signal-strength readings to a topic on the household message bus.

Each occupant has a personal device that is always with them — phone or wearable — broadcasting a known BLE identifier. The receivers see the broadcast, measure the signal strength, and report.

The orchestration layer subscribes to the readings and runs the trilateration calculation. The output is a probability distribution over rooms, updated every few seconds. The most likely room is the current presence of the occupant. Confidence below a threshold is reported as “uncertain”, and the rest of the system handles uncertainty gracefully.

The calibration that takes a weekend

The math gives you which receiver each broadcast is closest to. Translating that into rooms requires calibration, and calibration takes a weekend.

The procedure: walk through every room with the broadcasting device, dwell in each for a minute or two, log the receiver readings during the dwell. The result is a fingerprint per room: the expected signal strength to each receiver when the occupant is in that room. The trilateration calculation then matches new readings against the fingerprints.

The first calibration is wrong about ten percent of the time. The system gets recalibrated as we discover edge cases — sitting in a particular chair near the window, standing in a doorway, the bathroom which has unusually thick walls. Within a few weeks the accuracy is well above ninety percent, and the few percent of misclassifications are the kind that don't break automations because the relevant rooms are usually adjacent.

The automations that become possible

With reliable room-level presence, a long list of automations that previously required explicit configuration become possible.

• Lights follow the occupant. Dim or off in empty rooms; appropriate scene in occupied rooms; automatic transition when moving between rooms.
• Heating zones can be active where someone is and idle where no one is, with prediction of where someone is moving to.
• Music follows the occupant or stays put depending on configuration. Either is fine; both are now expressible.
• Notifications can be routed to the speaker in whichever room the occupant is currently in, rather than the whole house.
• The household briefing agent can use proximity to decide when to surface a piece of information, rather than waiting for an explicit query.

None of these is dramatic in isolation. The cumulative effect is a house that quietly tracks who is where and acts accordingly, which is the difference between a smart home and a cognitive domicile.

What the architecture deliberately does not do

The presence layer reports current location, with low-resolution history for the last few hours. It does not retain long-term location history, even within the house. It does not aggregate presence patterns into profiles. It does not share presence data with anything outside the local mesh.

The architectural commitment matters for the same reason the rest of the cognitive domicile is local-first: the privacy posture has to be invariant to who happens to control the orchestration platform at any given moment. We chose this architecture because we did not want to be in the position where a vendor change, a price change, or a policy change could turn our movement patterns into something visible to a third party.

The trade-off is that we cannot do the kind of long-term-pattern analytics that a cloud-based system could. We do not miss it. The capability we wanted is real-time, room-level presence, and that is exactly what the architecture provides.

The takeaway

BLE trilateration is the unglamorous infrastructure that makes most other household automations actually work. The hardware is cheap, the calibration is a weekend, and the resulting signal becomes the input to dozens of small behaviours that compound into a house that quietly does the right thing.

If your smart home today still relies on motion sensors and explicit schedules, the upgrade is worth doing. The architecture is local-first by construction, the privacy profile is acceptable, and the day-to-day experience is meaningfully better than what motion-based presence can deliver.