Physical AI data collection jobs
Physical AI is the name for models that have to deal with the real world rather than text on a screen: systems that recognise objects, follow what hands are doing, or help someone get through an ordinary day. They learn from recordings of that world. GoKuna pays adults to make those recordings at home, to a written brief, one task at a time.
Why physical AI needs people with phones
A language model can learn from text that already exists. A model that has to understand a kitchen, a cluttered shelf or a pair of hands folding laundry needs footage and sound of exactly those things, recorded in many different real homes, in different light, with different objects.
That variety cannot be staged in one studio. It comes from many people each recording a little of their own everyday surroundings, carefully and with consent. Collecting it is the job.
The recordings GoKuna pays for
Objects: one ordinary object on a clear surface, filmed steadily for 10 to 20 seconds from two sides. $5.00 per accepted clip, about 3 minutes.
Sound: a few seconds each of a kettle, a tap and a drawer, recorded where you normally cook, with no voices. $6.00 per accepted set, about 4 minutes.
Hands at work: short first-person clips of cooking, tidying or laundry, filmed with a phone or action camera mounted at head height. $10.00 per accepted clip, open worldwide.
A day in motion: longer stretches of everyday life from a head or chest mount, showing normal activity as it happens. $20.00 per accepted hour, in North America to start.
What good physical data looks like
Physical data is useful when it is ordinary and clear. The model needs to see the object whole, hear the sound without talking over it, and follow the hands without the camera swinging away. A brief spells out these details, such as clip length, angle, light and what must stay out of frame, because that is what makes one clip comparable with thousands of others.
You do not need to make anything look good. You need to make it look like it really is, and match the brief.
Consent and privacy come first
Recording the physical world means recording your own space. Only submit media you are authorized to share. Keep identifiable bystanders, private information and other people's property out of frame unless you have permission. Preview every recording before you upload it.
You keep ownership of what you record. When a recording is accepted, you grant a licence to use it for training and evaluating machine learning models, including on behalf of GoKuna's customers. You can delete your account at any time, which removes your profile and any recordings not yet accepted.
Review and payment
Every submission is reviewed by a person against its brief, usually within 24 hours, and the decision is emailed to you. Accepted submissions earn the reward shown when you reserved the task. Rejected ones come with the reason, and you can record again while the task is open.
Payouts go through Stripe to your own verified bank account, in over 100 countries. There is no minimum payout and no fee. Once a payment is sent, it usually reaches your bank within a few business days.
How to start
You need to be 18 or older with a phone that records video and sound. Sign in with your email address, confirm your age and country, and agree to only submit media you are authorized to share.
Open tasks and their rewards appear once you are signed in. Start with an object video or a kitchen audio set: both take a few minutes and teach you how briefs and review work before you try a first-person task.
What is physical AI?
Models that perceive or act in the real world, such as recognising objects, understanding sounds or following what hands are doing, rather than working only with text.
Do I need robotics or technical knowledge?
No. The tasks are recordings of everyday things. The brief tells you exactly what to capture.
Do I need special equipment?
A phone covers object, audio and handwriting tasks. First-person tasks need the phone or an action camera mounted at head or chest height, as the brief describes.
Why does consent matter for physical AI data?
Recordings of real homes can show real people and private things. Every GoKuna contributor agrees to only submit media they are authorized to share, and each task records which consent terms applied to it.
Can I record in any room?
Record where the brief says. Kitchen audio is recorded where you normally cook. Object videos need a clear surface. Household clips follow the activity wherever it happens in your home.
What if a recording is rejected?
You see the reason, and you can record a new one while the task is still open.
Is the pay guaranteed?
The reward for a task is fixed when you reserve it and paid when your submission is accepted. How much work is available is not guaranteed.