Sensors: motion, contact, temperature, leak and air quality

Motion is not occupancy#

The two words are used interchangeably on packaging and they describe different physics.

Passive infrared (PIR) watches for changes in infrared energy across a lens that splits its field of view into zones. When a warm body crosses from one zone into another, the signal changes and the sensor fires. Consequences fall straight out of that:

• A person who stops moving disappears. This is why the lights go off while you are reading on the sofa.
• Walking straight at the sensor is the weakest case, because you cross few zone boundaries. Mounting so people cross the field sideways works far better than mounting facing the door.
• Anything that moves heat triggers it: a heating vent, sunlight tracking across a floor, a radiator, a pet.
• It sips power. PIR is why a motion sensor can run for a year or more on a coin cell.
• It has a blind time. After firing, the sensor ignores everything for a cooldown period, often tens of seconds to a few minutes. That, not the radio, is usually what makes re-triggering feel slow.

mmWave radar transmits a radio signal and measures the phase shift of the reflection. Motion as small as breathing changes that reflection, so the sensor keeps reporting occupied while you sit perfectly still. The costs are equally structural:

• It needs continuous power, so these are mains-powered devices, and most of them are Wi-Fi rather than Thread or Zigbee.
• Radio passes through drywall. A radar sensor mounted on an interior partition will happily detect the person in the next room, and a wall is not the boundary you assumed it was.
• It reacts to non-humans: ceiling fans, curtains in an airflow, hanging plants, pets.
• Clearing can lag, because the sensor is deciding that a small signal has finally stopped.

The pattern that works in practice is both: a PIR for fast turn-on, an mmWave for hold-on, combined in the automation. Turn on when motion fires, stay on while presence is true, turn off on a timer after presence clears.

What Matter does not model#

The same limitation shapes what a fancy sensor is worth to you. Aqara's FP2, for instance, divides a room of up to about 40 square meters, roughly 430 square feet, into as many as 30 zones and tracks up to five people, with a built-in light sensor and an IPX5 rating. All of that lives in Aqara's app and its own integrations. Through a generic Matter path you get "occupied" or "not occupied".

There is also no vibration sensor and no tilt sensor device type, and no standard way to configure sensitivity or cooldown. Every one of those settings is vendor territory.

The other sensor classes, briefly#

Contact sensors are a reed switch and a magnet. They are cheap, reliable and boring, which is the highest compliment in this category. Two things go wrong: the published maximum gap between the halves is smaller than people assume, and steel doors and frames detune the antenna, so a sensor that works on a wooden door drops off a metal one. Mount the radio half on the frame rather than the moving door where you can.

Temperature and humidity sensors are wrong more often than they are broken. Placement dominates: an exterior wall, direct sun, a supply register or a spot above a warm appliance will all skew a reading by more than the sensor's own error. Combined sensors housed with mains electronics also self-heat. Most platforms let you apply a calibration offset, which is worth doing once against a reference thermometer.

Leak sensors are two probes or a rope-style extension cable. Placement is the whole product: put them at the low point where water will actually collect, not touching the appliance, because water runs downhill before it pools. Buy one with a local siren, because a push notification is worthless if your phone is on silent, and understand that a sensor does not stop water. Stopping water requires a valve. Note the version detail: water leak, freeze and rain detection arrived in Matter 1.3, not 1.2.

Air quality sensors are the least comparable category on the shelf. Particulate (PM2.5) sensors, volatile organic compound sensors and carbon dioxide sensors measure entirely different things, and many devices that display "CO2" are estimating it from a VOC reading rather than measuring it with an NDIR sensor. Chemical sensors need warm-up time, drift over months, and often self-calibrate against the cleanest air they have recently seen, which produces odd readings in a room that is never well ventilated. Air quality is a Matter 1.2 device type, and the certified population is small.

Battery life is the reporting interval#

This is the section that changes purchasing decisions.

A battery sensor is normally a sleepy end device: its radio is off most of the time, it wakes on a schedule to poll its parent, and everything addressed to it queues at the parent until it wakes. Matter formalizes this as the intermittently connected device model, with a short idle mode polling at most every 15 seconds and a long idle mode, added in Matter 1.3, that can stretch far further. Idle periods can run up to 18 hours.

Layered on top, controllers use subscriptions, not polling. A subscription has a minimum and a maximum interval, and the device must report at least once per maximum interval even if nothing has changed. That heartbeat is how a controller decides a device is offline.

So the dial has two ends and you cannot have both:

Sensible sensors report on change plus a heartbeat: temperature when it moves past a threshold, contact the instant it opens, plus a periodic "still here" message. The threshold is usually a vendor setting and rarely exposed through Matter.

Why "Thread battery sensors" sometimes disappoint#

Thread defines several end device types, and only a sleepy end device gets multi-year coin cell life. Several shipping battery sensors certify as full end devices with the radio always on, which is the real explanation for a sensor that should last two years and lasts four months. Aqara's own specification for one sensor claims about two years on Thread against three on Zigbee, which is the vendor acknowledging the cost.

Other documented drains:

• An unstable mesh, where the sensor constantly re-attaches to a parent.
• Too few Thread routers, so the sensor cannot hold a stable parent at all. Mesh design is the fix.
• Several ecosystems subscribing to the same node. Share a sensor across three fabrics and three controllers each keep their own subscription alive.
• Metal mounting surfaces detuning the antenna, which raises retransmissions.

Cold kills coin cells#

Sensors in a garage, a freezer or an outbuilding fail on a different timescale. In one independent freezer test, a generic CR2032 lasted about two days, cold-formula CR2032 cells lasted between a week and two months, and an 18650 lithium cell on the same sensor ran for over a year at -20 C. Both chemistries are around 3.2 V nominal. If a sensor is going somewhere cold, the cell type matters more than the sensor.

Living with sleepy devices#

Sleepy behavior explains a whole class of "broken" sensors that are working exactly as designed.

• You cannot poll one on demand. The app shows the last reported value, which may be minutes or hours old.
• Commands and configuration changes queue until the device next wakes. Most sensors have a button that wakes the radio immediately, and pressing it is the first step in almost every configuration task.
• Firmware updates need the device awake for a long stretch, which is why battery sensors are the slowest and least reliable devices to update, and why bridged sub-devices never get updates over Matter at all.
• "Unresponsive" usually means a missed heartbeat, not a dead device. Before replacing anything, check whether the parent router near it is still alive. Thread troubleshooting works through this in order.

Placement and false triggers#

• Mount PIR so people cross its field sideways. Room corners give the widest usable coverage, and facing a doorway head on is the weakest geometry there is.
• Keep PIR away from moving heat: supply registers, radiators, ovens, and any window where the sun tracks across the floor in the afternoon.
• Pet immunity is a lens and masking trick, not intelligence. It has a stated weight range and it is not reliable for a cat that climbs.
• Keep mmWave away from fans, curtains and airflow, and think about what is behind the wall you mount it on.
• Put leak sensors at the low point, and add a second one where the water would go next.
• Give temperature sensors an interior wall at seated height, away from doors and vents, then calibrate once.

Platform quirks worth knowing before you buy#

Alexa splits composed devices. A single sensor reporting temperature and humidity appears as two separate devices, because Amazon maps temperature measurement and relative humidity measurement onto separate capability interfaces. More awkwardly, Alexa silently drops endpoints carrying a cluster outside its expectation for that device type, so a multifunction sensor can lose one of its functions with no error message anywhere. If a sensor's third reading never appears, that is the mechanism, not a pairing failure.

Alexa also caps at 50 bridged devices per connection, and exceeding it makes devices vanish from the app rather than producing an error. A large Zigbee fleet behind one bridge hits that quickly.

Elsewhere, Home Assistant exposes the most sensor detail, including the diagnostics that make the battery and heartbeat behavior above visible rather than mysterious. Matter 1.6 added security sensor event history, which starts to close the gap between "the sensor fired" and "here is when it fired", though controller support is another question. What Matter still cannot do has the wider list.

How to choose#

• Motion for lighting in a hallway or closet: PIR on Zigbee or Thread. Cheap, fast, years of battery.
• Occupancy where people sit still: mmWave, mains powered, accepting Wi-Fi and the vendor app.
• Doors and windows: contact sensors, checking the published gap against your actual door.
• Water: leak sensors with a local siren at every low point, and a valve if the risk is real.
• Anything cold or outdoors: check cell chemistry and operating temperature before the radio.