Presence-Based Automations: Knowing Who's Home
Your phone says you left while you're on the couch. Motion sensors miss you when you're still. Here's how to combine presence signals that actually work.
Prices in this article are typical U.S. retail, checked October 2026.
Section 1: Why doesn’t the house know you’re home? (Introduction)
Presence detection is a guess built from signals: phones, motion, door codes and manual modes. Combine at least two, accept that the guess will sometimes be wrong, and give the household a manual override that everyone knows how to use.
The away routine fires while you are sitting on the couch. The bathroom light turns off while someone is reading. The hallway light turns on for the dog and off again before you reach the door. None of these are hardware failures. They are presence failures: the house is guessing whether someone is there from a signal that does not actually mean that.
Presence is the hardest input in home automation because there is no single sensor that answers “is a person here?” A phone knows where it is, not where you are. A motion sensor knows that something moved, not that someone stayed. A door sensor knows the door opened, not who came in or whether they left.
The fix is not a better sensor. It is a confidence model: combine two or three weak signals, add time, and only then change the house’s state. This guide covers the four signals worth using, how each one fails, the logic that combines them, and the setups where the extra effort pays off. For the wider automation picture, start with the Automation hub.
Section 2: What signals can tell the house you’re there?
Most homes already have two of these, and the other two are inexpensive.
| Signal | What it actually knows | Latency | Fails when |
|---|---|---|---|
| Phone location | Where the phone is, roughly | Seconds to minutes | Battery saver, GPS drift, phone left at home |
| Motion (PIR) | Something warm moved in the last few minutes | Instant | You sit still, a pet walks by, a vent moves a curtain |
| Presence radar (mmWave) | Micro-motion, including breathing, in a zone | Seconds | You are out of range, behind a wall, or the room is too large |
| Access events | A door opened, a code was used, a lock changed | Instant | It tells you about the door, not the person |
Phone location
Geofencing is the default presence signal because every phone has it. It is also the least reliable at the boundary: GPS drifts, phones sleep, and a phone can be at home while its owner is not. Google’s presence-based automations explicitly rely on phones that home members opt in, and they acknowledge that the signal comes from compatible devices rather than from certainty.
Using it well means accepting its limits: a geofence should be generous, and “away” should require the phone to be gone for several minutes, not one.
Motion (PIR)
A passive infrared sensor detects a change in heat across its field of view. It is fast, cheap, and local. It is also a motion detector, not an occupancy detector: sit still for the timeout period and it reports the room empty. The Philips Hue Motion Sensor at $48.99 is a good example of the category — reliable, battery-powered, and best used for “someone just walked in,” not “someone is here.”
Presence radar (mmWave)
Millimeter-wave radar detects micro-motion, including the small movements of breathing. That is what makes it a presence sensor rather than a motion sensor: a person reading on the couch still shows up. The Aqara Presence Sensor FP2 at $82.99 is the best-known example, with zone positioning across a room up to about 430 square feet and detection of up to five people.
It is not magic. Radar is blocked by walls, confused by some ceiling fans, and range-limited; the FP2 is wired, which is a real constraint in a rental or a room without an outlet near the ceiling.
Access events
A smart lock, keypad, or door sensor is an event signal: the front door opened at 6. It is excellent for triggering an arrival scene and useless for knowing whether the person is still there. Combine it with a phone or a radar, not instead of one.
A camera can also detect people, and it is the most accurate signal of all. It is also the most privacy-invasive, the most expensive to run continuously, and the one that fails in the dark or when the lens is blocked. For presence, a camera is the wrong default; the sensor directory covers where it fits.
Section 3: How does each signal fail?
The failure modes are predictable, which is good news: you can design around them.
Phone location fails at the edges. A phone that drops off Wi-Fi can look “away” while it sits on the nightstand; a phone that sleeps aggressively can delay the geofence event by minutes. iOS and Android both manage background location to save battery, and the automation platform only knows what the OS reports.
PIR fails when nothing moves. It also fires for pets, sunlight through a window, and heating vents. A motion sensor pointing at a doorway is reliable; one pointing across a living room is a random number generator.
mmWave fails at range and through obstructions. It can also struggle with multiple people in overlapping zones, and it needs power. The FP2’s own documentation is careful to say it is not a medical device, which is the right frame: it is a presence sensor, not a health monitor.
Access events fail because a door is not a person. A door can open and close with nobody entering; a window can be left open; a guest can use a code and leave an hour later. The event is a fact, not an occupancy state.
The household is the fifth failure. Roommates, guests, and children do not necessarily carry the phone that the automation is tracking. A presence model that only knows the owner’s phone will be wrong every time someone else is home.
Two more signals exist and are worth knowing. Bluetooth beacons and key fobs can mark a person or a set of keys, but they need batteries and a receiver nearby. Router presence — the phone’s MAC address on the Wi-Fi — is fast and local, but it fails the moment the phone sleeps or the Wi-Fi drops. Neither replaces the four above; both can be an extra vote.
Section 4: How many signals should it take to say you’re away?
The practical approach is to treat presence as a vote, not a fact.
Require two signals for “away”
Going away is the high-stakes decision: it turns off lights, sets the thermostat, and arms the alarm. Require the phone to be away and no motion for a dwell period — five to ten minutes is typical. One signal alone should never arm the house.
Coming home is the low-stakes direction, so it can be more permissive: a phone crossing the geofence or a door code or motion after a long absence is enough to set “home.”
Add time, not more sensors
Hysteresis is the technical name for not flipping on the first sample. A state machine with three states covers most homes:
- Home: any trusted signal says someone is present.
- Away: phone away for X minutes and no motion for X minutes.
- Guest: a manual override that ignores the owner’s phone.
The guest state matters. A house that arms itself because the owner left, while a guest is still inside, is a false alarm waiting to happen.
| State | Enter when | Leave when |
|---|---|---|
| Home | Any trusted signal says present | Everyone is away for the dwell period |
| Away | Phone away and no motion for 5–10 minutes | A door code, geofence entry or motion |
| Guest | A physical button or app override | The same button, or the guest leaves |
The table is not a product feature; it is the logic you implement in whichever app you use. Write it down once and the individual automations become obvious.
Use entry and exit differently
An entry event (a code, a door sensor) is a good trigger for “arriving” and a bad trigger for “away.” Exit events are the opposite: the door closing is a hint, not proof. Combine the hint with the phone.
Test the boundaries
Walk out with the phone and come back within two minutes. Sit still for twenty minutes in a room with a PIR. Put the phone in airplane mode. The failures you find in a test are the ones that will not surprise you at 2 a.m.
Platforms help here. Home Assistant’s person integration combines device trackers — a phone, a router, a GPS source — into a single person entity, which is exactly the voting model described above. Google Home’s presence settings do the same at a coarser level, with phone location and compatible devices.
Section 5: How do you combine the signals in practice?
Arrival lighting. Trigger on a door code or a phone geofence crossing, then turn on the entry light only if it is dark. The phone handles “coming home”; the lock handles “someone actually opened the door.” Two signals, one scene.
Away mode. Require the phone away and no motion for ten minutes. Add a delay before arming the alarm so the door can close and the person can walk away. Do not arm on a single geofence event.
Room-level presence. An mmWave sensor in an office or bathroom keeps the light on while someone is still. This is where radar earns its price: a PIR cannot do it. The FP2’s zone feature lets the same sensor ignore the doorway and watch the desk.
Elder check. A presence sensor can tell you that a parent got out of bed and moved around the house, which is useful context for a family check-in. It cannot tell you whether they are well. Keep the language careful; a presence sensor is not a medical device.
Guest mode. A physical button or a virtual switch that disables away automations while a guest is in the house. It is the simplest fix for the household failure mode, and it costs nothing.
Roommates. If two people share the flat, the presence model needs both phones, or a rule that treats any phone at home as “home.” A single-phone model will turn the lights off on the other person. Add the second phone before adding a radar.
Section 6: What is the three-signal plan? (Conclusion)
- Start with the phone and the door. A geofence plus a door sensor or lock event covers arrival and departure for one person, with the dwell time that prevents flapping.
- Add a presence sensor only for the room that needs it. The office, the bathroom, or the living room where someone sits still. One radar sensor in the right room beats five PIRs everywhere.
- Add a guest override before you trust away mode. A button that says “someone is here” is the difference between a smart house and a house that arms itself with a guest inside.
The automation audit is the companion exercise: once presence is a state, the rules that read it can be checked for conflicts like any other.
Some homes should skip this. If you live alone and your phone is reliable, a generous geofence plus a ten-minute dwell is enough. If you have pets and PIR sensors, radar is the upgrade that stops the false triggers. And if you have a monitored alarm, let the alarm’s own entry/exit delays do the arming; do not build a second presence model on top of it.
Presence is not a sensor you buy. It is a conclusion you earn by combining weak signals and giving them time. The house does not need to know exactly where you are. It needs to be right about whether you are there.
Where to buy
Opens an Amazon search. As an Amazon Associate, we earn from qualifying purchases. No product in this guide was supplied or tested by a manufacturer.
Related reading
- The Automation Audit — finding rules that fight each other.
- Set-and-Forget Automation — designing routines that hold up.
- The Smart Sensor Directory — motion, presence and contact sensors compared.
- Automation Fail-Safes — what works when the network does not.
Sources
- Presence Sensor FP2 — Aqara US store, price, zone coverage and multi-person detection checked October 2026
- Hue Motion Sensor — Philips Hue, price and PIR specifications checked October 2026
- Change presence-based automations settings — Google Home Help, phone location and compatible devices for Home & Away checked October 2026
- Person — Home Assistant, combining device trackers into a person entity checked October 2026