A humanoid is not a home security system
The search that lands people here is usually some version of “can a robot guard my house.” The honest answer is no, and it is no in a specific way: detection, recording and 24/7 coverage are solved problems, and they are solved by cameras, door sensors and an alarm that costs a few hundred dollars, not twenty thousand. A 2026 humanoid loses to that stack on every axis that actually prevents a break-in — field of view, battery, silence, always-on recording, and a professional monitoring path.
What a humanoid can add, once that stack is already in place, is a mobile viewpoint. Something on the network triggered. The robot walks over, looks, and you decide whether it was the cat, a branch, or a person. It can also occupy the house while you are away in a way a timer on the lights cannot, and it can do physical checks a camera cannot: is the back door actually latched, is a window left open, did someone leave a garage door off the catch. Those are real jobs. They are not “home security” in the sense the query usually means, and treating them as a replacement for an alarm is how this purchase goes wrong.
What it can actually do in 2026
Investigate a trigger. This is the one job where a humanoid is not a worse camera. A fixed camera sees one slice of a room; a robot can be sent to the slice that just changed. In a house with a real alarm, that is the difference between driving home at 1 a.m. and looking at a 20-second clip. It only works if the robot can be dispatched from your phone, if it can navigate in the dark, and if it still functions when the home internet is degraded — more on that below.
Physical checks. Lock the doors, close a window, switch off a stove you are not sure about, look in on a room a camera does not cover. These are household tasks with a security flavour, and they are within what consumer humanoids actually do. They are also slow, and they need a mapped home — our mapping guide is a prerequisite, not an optional extra.
Occupancy. A robot moving through a house, talking, turning lights on, is a more convincing “someone is home” signal than a lamp on a timer. It will not fool a determined burglar who has watched the house. It will change the odds on the casual case, which is most of them. Treat it as a supplement to lights and a visible alarm, not as a strategy on its own.
Escalation. Some platforms can place a call, sound an alarm, or notify a nominated contact when they see a person they do not recognise. That is useful. It is also the same job a $150 camera already does, usually with better night vision and without needing a charged robot standing in a hallway.
What it must never be asked to do
Confront an intruder. No consumer humanoid is rated for that, no manufacturer supports it, and a 20-kilogram machine walking toward a person who may be armed is a way to get a robot destroyed and a household endangered. Configure this limit explicitly. The failure mode is not the robot doing it badly — it is a family member assuming the robot “handles” a break-in so they do not need to call the police.
Replace the alarm. A humanoid patrols a route, then it charges. Cameras do not. An alarm panel does not. If the robot is the only thing watching the house, the house is unwatched for the hours it spends on the dock, and unwatched the moment a firmware update, a mapping error or an empty battery takes it offline. Keep the alarm. Keep the cameras. Keep the monitoring subscription if you have one.
Be the evidence. A robot’s onboard video is not a substitute for a camera system with cloud or local NVR retention, timestamps and a chain of custody you can hand to an insurer or the police. Our insurance guide covers what actually gets paid out after a household incident; robot footage is supporting colour, not the policy’s idea of proof.
| Job | Humanoid role | What actually does it |
|---|---|---|
| Detect a break-in | Secondary, slow, battery-limited | Cameras + door/window sensors + alarm |
| Record evidence | Incidental clips | NVR / cloud cameras with retention |
| Stop an intruder | None | A human, the police, or a locked door |
| Investigate a trigger | Primary — this is the useful job | You, on a camera app, if you can see enough |
| Check a door or window | Primary | You, in person |
| Occupancy while away | Useful supplement | Lights, timers, a visible alarm bell |
| 24/7 watch | None — it charges | Cameras |
Which robots are worth considering for this job
The scores that matter here are not capability in the household-chore sense. Privacy decides whether you can tolerate a robot watching an empty house. Reliability decides whether the dispatch you send at 1 a.m. actually happens. Ease of use decides whether anyone in the family can send it without you. On-device processing decides whether a cut internet connection also cuts the only thing walking the hallway.
| Robot | Price | HOSI | Privacy | Reliability | Capability |
|---|---|---|---|---|---|
| 1X Neo | $20,000 | 9.0 | 9.2 | 8.7 | 8.9 |
| Figure 03 | $20,000 | 9.2 | 9.0 | 9.1 | 9.4 |
| Tesla Optimus Gen 3 | $20,000 | 8.5 | 8.0 | 8.2 | 9.0 |
| Unitree G1 | $16,000 | 7.7 | 7.3 | 7.6 | 8.0 |
1X Neo is the default for this job, even though Figure 03 scores higher overall. A security robot is a camera you invited inside, and Neo is the only consumer unit that combines a 9.2 privacy score, on-device speech and navigation, and a hardware microphone and camera cut-off you can operate without opening an app. When the internet drops, the reminders and the map keep working. That is the difference between “the robot looked” and “the robot was a sculpture in the hallway while something happened.”
Figure 03 is the more physically capable option at 9.4, with the best reliability score in the consumer tier (9.1). If the job is mostly physical checks — doors, windows, a garage — rather than leaving a robot as a watcher, Figure is the one. Privacy at 9.0 is still in the top band.
Tesla Optimus Gen 3 can do the physical work. Privacy at 8.0 is the weakest of the $20,000 cluster, and for a device whose job is to look at an empty house that is a real objection, not a spec-sheet footnote. Unitree G1 at $16,000 is the budget door. Privacy 7.3 and reliability 7.6 are below the line we are willing to recommend for a robot left alone in a home; it belongs in the table so you can see the trade, not so you buy it for this job.
We do not shortlist presence-first machines (Sunday Companion, Neura MIPA) here. They score 6.5–6.6 on capability and were not built to walk a house in the dark. We also do not shortlist $40,000–$80,000 platforms that appear in “security robot” roundups; those are commercial or research machines, not household purchases. The home security hub now follows this same consumer-only rule.
The attack surface you are adding
A humanoid in security mode is a networked camera with legs, a microphone, a map of your house and, on most platforms, a cloud account. That is a larger attack surface than a $200 camera, not a smaller one. If you would not hang an always-on webcam in the hallway without knowing where the footage goes, do not skip that question for a robot that can also walk into the bedroom.
Run the checks in our privacy audit before you ever set a patrol routine: local vs cloud processing, retention, who can view a live feed, and whether the hardware cut-off actually cuts power. Then test the outage case. Unplug the router. Send the robot to the back door. If it cannot go, it is not a security device — it is a cloud appliance that happens to have legs.
What it costs, and what else you should buy first
A $20,000 robot models out near $32,300 over five years — about $538 a month — once warranty, service, a spare battery, repairs and the insurance delta are counted. The full model is in the five-year cost breakdown; financing changes cash flow, not the commitment.
Put that number next to a real security stack. A competent camera set, door sensors, a monitored alarm and a spare lock job on the back door will typically land well under $2,000 in the first year, often under $1,000. If you do not have that stack, spend there first. The humanoid does not catch up on detection no matter which model you pick. It starts to make sense only as the thing you send after the stack has already told you something moved.
How to set it up so it is still useful in month six
Map the house properly, including night lighting, glass, stairs and the route to every exterior door. Leave no-go zones on bedrooms during sleeping hours and on bathrooms always — a “security” robot that walks in on someone at 2 a.m. gets unplugged, and then it is not securing anything. Keep motion speed conservative and make the robot audible; a silent machine in a dark hallway is how you frighten the people who live there, which is the opposite of the job.
Start with one routine: “when the kitchen camera flags, go to the kitchen.” Run it for two weeks. Adding a full-house patrol on day one is how you discover the docking problem, the rug problem and the “it got stuck behind a chair” problem all at once, at night. Robot-proof the home the way you would for any other job — charging location, doorways, clutter — because a security dispatch that cannot fit through the utility-room door is theatre.
Agree who gets the live view. Unannounced family access is how trust dies, and it is also how a “security” setup becomes domestic surveillance. Write down who is notified, who can dispatch the robot, and what happens if the robot sees a person it does not recognise — call a nominated human, not “handle it.”
Who this is actually for
This is for a household that already has cameras and an alarm, wants a mobile viewpoint for the alerts those systems cannot see clearly, and is willing to administer a robot. It is not for someone hoping to skip the alarm panel. It is not for a landlord who wants to watch tenants. It is not a substitute for a dog, a neighbour, or a lock.
If the robot would mostly be company, that is a different job — the companion buying guide. If you are not sure which job you are shopping for, the Robot Finder asks about the home and narrows it in a couple of minutes.
