Mohamad Dunggio is 23 and lives in Gorontalo, on the Indonesian island of Sulawesi. Since April he has been earning extra money in a way that would have sounded like a joke two years ago: he straps a phone to his head and films himself doing housework. Washing dishes, ironing, making the bed, folding clothes. One session of dishwashing ran four hours, long enough to leave his fingers wrinkled.
Three companies, based in Argentina, India and the United States, paid him between $3 and $10 an hour for it.
Somewhere in the US, someone is doing exactly the same thing with a five-camera rig on their head, chest and wrists, booked through a staffing app, and earning $13 to $25 an hour, occasionally up to $50.
They are both doing the same job: teaching humanoid robots how a human hand picks up a cup. And it is now one of the stranger, more accessible side incomes of 2026. Here is what it is, what it really pays, and what the ads leave out.
What the job actually is
Robots that are supposed to work in homes have a data problem. A chatbot learned language by reading the internet. There is no internet of hands folding towels. So robot companies are buying it, one recorded hour at a time.
The industry calls it egocentric data: video and sensor recordings captured from the point of view of the person doing the task, with their hands in frame. Researchers use it roughly the way language models use text, to learn what usually comes next. Grip the plate here, turn it this way, put it there.
| What you do | What the company wants |
|---|---|
| Wear a camera on your head or chest, sometimes on your wrists too | footage from your point of view, hands always visible |
| Do ordinary chores: dishes, laundry, sweeping, making coffee, watering plants | variety: different kitchens, different hands, different objects |
| Follow a script of tasks | consistent lighting, no reflections, nothing skipped |
| Upload the footage | hours that pass quality checks |
The scale is already large. Appen, an Australian-listed data company, says it can collect about 60,000 hours of video a month from up to 5,000 contributors across countries including India, Canada, the US, Brazil, Mexico and the Philippines. Instawork's robotics lab has estimated the industry would collect 20 million hours of recordings this year. DoorDash launched a Tasks app in March that pays some of its drivers to film chores, notably leaving out states with strict data privacy laws.
What it actually pays
This is where the headlines and the reality drift apart, so here is the full range rather than the best number.
| Where and how | What people report being paid |
| Indonesia, paid by overseas data companies | $3 to $10 an hour |
| China, a stay-at-home mother filming chores | about $3 an hour (20 yuan) |
| Contractors in Kenya, the Philippines, India, Brazil via Micro1 | around $15 an hour |
| US, via staffing apps such as Instawork, gear supplied | $13 to $25 an hour, some recordings up to $50 |
| One US worker, a single session | about $80 for two hours of usable video |
| US teleoperation (remotely driving a real robot) | $25 to $65 an hour in checked job postings |
The same hour of dishwashing is worth between $3 and $50 depending on which passport the hands belong to. That gap is the whole economics of this industry, and it is also, unexpectedly, good news if you live somewhere wealthy, which I will come to.
What a realistic month looks like
Forget the $50 headline and work it through the way a normal person would.
Say you are in the US, you pick up two recording shifts a week through a staffing app, each about two hours, at $20 an hour, and roughly 85% of your footage passes the quality check.
| Per week | Per month | |
| Hours recorded | 4 | ~17 |
| Gross at $20 an hour | $80 | ~$347 |
| After ~15% rejected footage | $68 | ~$295 |
That is a little under $300 a month, which lands almost exactly on the median of what people make from AI side work generally. I looked at that distribution in what 79 million Americans actually earn with AI, and the answer was the same shape: real, modest, and nowhere near replacing a salary.
Double the shifts and you are at around $600. To get to the $50-an-hour end you need the specialist sessions, the multi-camera rigs and the projects that want particular homes or skills, and those are the first to fill.
The catches nobody puts in the ad
Rejected footage is unpaid footage. Hands drift out of frame, the lighting changes, a shiny pan reflects the camera, you skip a step in the script. Each of those can sink a session. Dunggio had videos rejected. The pay is per accepted hour, not per hour of your life.
Getting your money out can cost money. Dunggio's specific complaint was that some platforms charge about $1 to withdraw balances under $10. At $3 an hour, that is a third of an hour's pay. Contributors on worker forums also report late payments, slow support and accounts closed without explanation.
You are filming the inside of your home. That is the product. Your kitchen, your possessions, sometimes your family walking past. A Monash University robotics professor, Dana KuliΔ, flagged exactly this: whether contributors understand how their data will be used, the privacy risk of recording inside homes, and footage being resold to third parties. Appen says it lets contributors ask for their data to be withdrawn even after it has been sold. Other platforms do not say.
You probably do not own what you record. Several reports note that contributors get flat hourly pay and keep no equity, royalties or ongoing rights to their footage. The recording you made today can train products sold for years. You are paid once.
Read the terms before the first shift. Who owns the video, whether it can be resold, whether faces and family members are blurred, how to request deletion, what counts as a rejection, and when you get paid. If a platform cannot answer those five questions, it is not a platform worth filming your kitchen for.
The good news if you live somewhere expensive
Here is the twist that turns this from a story about cheap labour into an opportunity for Western readers.
Appen says it is shifting collection toward wealthier countries like the US and Canada, because bulk footage from lower-cost countries has become commoditised and less valuable. The reason is almost comically specific, and its vice-president of delivery put it in one line: "The dishwashers in the US look very different from the dishwashers you would have elsewhere."
A robot that will be sold into American homes has to learn American kitchens. That means the footage that is scarce, and therefore better paid, is footage from the homes where the robots will actually be sold. Your ordinary kitchen, the one you think is unremarkable, is exactly what the training sets are missing. The same logic is why the robots themselves are priced the way they are in the humanoid price comparison: the expensive part is making them work in real homes, not building them.
The better-paid version: driving the robot yourself
Above filming chores sits a second tier, and it pays like a proper job.
Teleoperation means remotely controlling an actual robot, usually with a VR headset and hand controllers, to show it how to do a task so it can learn from the demonstration. Postings checked by one robotics job board ranged from $25 to $43 and $26 to $65 an hour, and earlier listings for 1X's humanoid VR operators sat around $25 to $35 an hour. Tesla advertised its Optimus data-collection role at up to $48 an hour back in 2024, with operators in motion-capture suits.
The trade-off is that these roles are usually on-site in a lab, often part-time contract work, and some marketplaces require 8 to 40 hours of hardware certification before your first paid shift. It is a real entry point into robotics for someone without an engineering degree, which is rare. It is also the kind of job that exists because the robots cannot yet do the task on their own, a gap I covered in why the robot brain is harder than the body.
Who this is actually for
Good fit: you want a few hundred dollars a month, you are fine with being on camera, you live in a market where staffing apps run these gigs, and you will read the terms. Students, people between jobs, anyone who already does side shifts on apps.
Good fit, higher tier: you are curious about robotics, you can commit to part-time lab shifts, and you would like a foot in the industry. Teleoperation pays far better and teaches you something sellable.
Poor fit: you need replacement income, you are uncomfortable recording your home or family, or you are looking at a platform that will not tell you who owns your footage.
If you want income that grows rather than an hourly rate that stays flat, this is a starting point rather than a destination. The routes that compound are in the nine ways people actually make money with AI, and the cheaper entry points into the robot business itself, such as buying a robot dog and renting out its work, are a different kind of bet.
Am I training the robot that replaces me?
It is a fair question, and the honest answer is: not soon, and probably not the way you fear.
The researcher quoted above put the gap precisely. There is a big difference between a robot folding laundry in a demonstration and doing it reliably every day for months, and as she said, "that last 1 per cent is taking much more time than anyone initially thought it would." Cost, maintenance and repair are still barriers. The home robots that exist today, such as 1X's Neo at $499 a month, still lean on remote human operators for the hard parts.
What is actually happening right now is the reverse of replacement. Robot companies need far more human hands than they have, and they are paying for them by the hour. Whether that lasts five years or fifteen, it is a strange new door that is open today.
Frequently asked questions
Can I really get paid to film myself doing chores?
Yes. Robot companies and data firms pay people to record household tasks from their own point of view, with hands visible, to train home robots. In the US this is often booked through staffing apps such as Instawork, which supply the camera gear.
How much does filming chores for robot training pay?
In the US, roughly $13 to $25 an hour, with some recordings reported at up to $50 an hour and one worker earning about $80 for two hours. In lower-cost countries the same work pays $3 to $15 an hour.
What equipment do I need?
Usually none of your own for app-based US gigs, which supply the gear, sometimes a five-camera rig on the head, chest and wrists. Overseas platforms often ask you to use your phone mounted on your head or chest.
Do I own the videos I record?
Usually not. Several reports describe flat hourly pay with no equity, royalties or ongoing rights. Check the contract for ownership, resale, blurring of faces and deletion before you start.
What is robot teleoperation and what does it pay?
Remotely controlling a real robot, usually with a VR headset, so it can learn from your demonstration. Checked US job postings range from about $25 to $65 an hour, typically for part-time, on-site lab work that may require certification first.
Sources: ABC News: egocentric training's gig workers; Komando: robotics companies pay people to film chores; The Washington Post: robots are learning housework from videos of humans; Rest of World: China's human labour behind humanoid robot data; robottraining.jobs: operator roles and advertised pay.
