Key Takeaways
- Every sensor type has its own failure mode when used alone — sensor fusion cancels these out by requiring agreement
- Built with standard Home Assistant automation conditions (multi-condition triggers or template sensors), not special hardware
- Common pattern: radar + PIR motion for occupancy, contact + camera for security events
- For several imperfect signals, Home Assistant's Bayesian binary sensor weighs each one by reliability instead of using fixed AND/OR rules
- The trade-off is added automation complexity and slightly more setup time for meaningfully fewer false triggers
- See the radar presence sensing guide for the specific sensor type most commonly fused with motion detection
Why One Sensor Isn't Enough
Every sensor type has a specific blind spot: PIR motion sensors miss stationary occupants, radar can occasionally false-positive on appliance vibration, contact sensors don't identify who triggered them, and cameras raise privacy considerations cameras-free sensors avoid.
- A PIR-only occupancy automation will turn off lights on someone sitting still, a well-known annoyance with PIR-only smart-home automations.
- A radar-only automation can occasionally register a false positive from a fan, HVAC vibration, or a pet, depending on sensitivity settings.
- Relying on any single sensor means inheriting that sensor's specific failure mode across every automation built on it.
Building a Fusion Automation
In Home Assistant, sensor fusion is typically built as a template binary sensor that combines two or more source sensors' states with AND/OR logic, which downstream automations then treat as a single, more reliable trigger.
- A basic occupancy-fusion template sensor might report "occupied" only when either the radar sensor is active, or the PIR sensor triggered within the last few minutes — covering both instant detection and the stationary-occupant case.
- For security automations, combining a contact sensor (door opened) with a camera-based person-detection event (see the local AI security camera guide) before triggering a notification reduces false alarms from, for example, a door opening with no one visible.
- Keep the fusion logic in one template sensor rather than duplicating the multi-condition logic across several separate automations — this makes it easier to tune later.
Common Fusion Patterns
The most common fusion patterns are occupancy (radar + motion), security (contact + camera detection), and environmental (multiple temperature/humidity sensors averaged to smooth out a single sensor's placement bias).
- Occupancy: radar sensor for stationary detection, PIR for instant motion response, combined so lighting automations react quickly and don't time out on someone sitting still.
- Security: door/window contact sensors combined with camera-based person detection, so a notification only fires when both a physical entry event and visual confirmation align.
- Environmental: averaging multiple temperature sensors in a room to reduce the impact of one sensor being placed in a draft or near a heat source, giving climate automations a more representative reading.
Weighted Fusion and Robustness
When simple AND/OR rules are not enough, Home Assistant's Bayesian binary sensor weighs sensors probabilistically: each observation moves a probability up or down, and the sensor turns on once it passes a threshold.
- Bayesian binary sensor: you set a prior (the baseline probability), then give each observation (a state, numeric range or template) a prob_given_true and a prob_given_false. The sensor turns on when the resulting probability exceeds probability_threshold (default 0.5). It can be set up in the UI (percentages) or in YAML (values from 0 to 1).
- Choose fixed template logic when the rule fits in one sentence (radar OR a recent PIR trigger). Choose the Bayesian sensor when several imperfect signals — radar, PIR, a light switch, a phone's presence — each carry partial evidence. You estimate the probabilities yourself, so start simple and tune.
- Add a delay off to the fused template binary sensor so brief gaps between sensor reports do not flip a room to unoccupied. Template binary sensors support delay_on and delay_off in YAML and under Additional options in the UI helper.
- Handle offline inputs deliberately: a template entity is always available unless you configure an availability template, so an offline source sensor simply fails its 'on' test. Check source availability in the template if a missing input must not be read as 'no activity'.
- For environmental fusion, the Min/Max helper combines two or more sensors that share a unit into a mean, median, minimum or maximum value (UI helper or YAML).
Frequently Asked Questions
Do I need special hardware for sensor fusion?
No — it's an automation-logic pattern built with standard Home Assistant template sensors and multi-condition triggers, combining whatever sensors you already have.
What's the simplest sensor fusion example to start with?
Combining a radar presence sensor with a PIR motion sensor for occupancy detection is a common starting point — it directly addresses PIR's "stops detecting a stationary person" limitation.
Does sensor fusion add noticeable automation complexity?
Some — you're writing a template sensor with combined logic instead of pointing an automation at one sensor directly. This is a manageable trade-off for meaningfully fewer false triggers on automations that matter (like security notifications).
Can I fuse a local LLM's reasoning with sensor data?
Yes — once fused sensor data is exposed as a standard Home Assistant entity, an LLM automation (see the AI automations guide) can reason over it the same way it would any other sensor, potentially adding further context-aware logic on top.
Is this the same as a home digital twin?
Related but narrower — sensor fusion combines a few sensors for one specific automation decision. A digital twin (see that guide) is a broader, ongoing model of the whole home's state that many automations could draw from.
When should I use the Bayesian sensor instead of a template sensor?
Use a template sensor when a simple AND/OR rule captures the logic. Use the Bayesian binary sensor when several imperfect signals each give partial evidence and you want them weighed by reliability — it needs a prior and per-observation probabilities that you estimate yourself.
What happens to a fused sensor when one input sensor goes offline?
By default a template sensor stays available and an offline input simply fails its condition, so the fused result can be wrong without any warning. Add an availability template or check the input states if the result must not be trusted while an input is missing.