Core modality · 01
Egocentric video
First-person footage from a head-mounted camera. What the worker sees is what the model sees — the hands, the tools, the part, every step.
Marxen collects real work video to teach robots how physical jobs are done. Below is the complete taxonomy of the work we cover — 26 segments, 257 venues, 808 scenes and 234 job roles — and every one of them is rated for fit.
Search the work your people actually do. You will get a straight answer, including a no.
26 segments · every one rated
We are collecting here now. If your floor looks like this, we can start with a paid trial.
A real fit, scheduled in waves. Send photos of your floor and we will size it with you.
Not a priority right now — either the work is mostly screen, speech and people, or consent cannot be cleanly obtained.
Your work does not map cleanly to the taxonomy yet. Send photos and we will tell you honestly.
Core modality · 01
First-person footage from a head-mounted camera. What the worker sees is what the model sees — the hands, the tools, the part, every step.
Core modality · 02
A wide lens set so both hands stay inside the frame through the whole task. This is what lets a model learn two-handed coordination rather than a single gesture.
Core modality · 03
Continuous recordings of a job from start to finish, not clips. The order of steps is as much of the signal as the steps themselves.
Core modality · 04
Every hour lands tagged against this taxonomy — segment, venue, scene, job role — so a lab can draw exactly the slice it needs.
Layers we can add on request
A second angle on the same task, time-synced to the head camera.
Motion and orientation streams alongside the video.
Spatial capture where a programme calls for geometry, not just pixels.
The worker talking through what they are doing, in their own language.
Built to a lab's own model-training spec, on the same network.
01
The work and both hands stay inside the camera view. Hands drifting below the frame for long stretches is the single most common reason an hour fails.
02
Clear, steady, well-lit footage of the task. Not the ceiling, not the floor, not a colleague across the room.
03
The worker's ordinary job at ordinary speed. Nothing staged, nothing performed for the camera, no repeating a task to fill time.
04
Whole sequences beginning to end. A robot learns from the full arc of a job — pick up, position, work, finish, set down.
05
Strap level on the forehead, lens pointed at the work. Our team fits it and trains each worker on day one.
06
No faces held in frame, no personal or private areas, no confidential documents or screens. Those segments get cut, and cut time is not paid.
This assumes every recorded hour clears our quality check. It will not in week one — straps slip, hands leave the frame, a camera gets put on backwards. Expect the first week to land lower and to climb once your workers are used to it.
Your factory could earn
₹62,500
per month
The line runs exactly as it runs today. The camera is the only new thing on the floor.
Our technical team brings the cameras, fits them, trains your workers, and handles the upload.
A small paid trial first. If the hours come out clean, we scale it up with you.
A short call to understand the work your factory does.
Our team comes to your factory and brings the cameras.
We fit the head strap and show each worker exactly what to do.
The worker does their normal job. The camera records. Nothing about your line changes.
At the end of the shift, the worker returns the camera to our team.
We upload through our own Marxen app. Safe, private, and traceable.
Footage is uploaded through our own Marxen app, not a consumer cloud drive. See the privacy policy for how it is stored and who can reach it.
Over and above their normal wage, only for the days they wear the camera.
When there is something in it for them, they wear it properly and willingly.
A camera worn properly is the difference between a paid hour and a rejected one.
No worker is forced. Taking part is always their decision, and they can stop at any time.
What to do next