Zirconoid is organizing human-captured data for physical AI

Physical AI is the future. Zirconoid provides global, human-captured egocentric data to train physical AI models at scale.

Our core belief is that complex data required to teach today's models is bottlenecked by access to diverse, real-world physical environments. Engineering systems that solve technical problems get commoditized. Proprietary human and operator-collected data does not.

From factory floors and fields to construction sites and manufacturing plants, we recruit global talent and collect large-scale datasets across diverse physical environments.

Trusted by the data providers who support the frontier

Our work

Datasets collected by real people

Textile manufacturing
Egocentric view of a textile worker inspecting fabric on a loom in an Indian mill

Egocentric video from textile factory floors

Factory workers wear head-mounted cameras through loom operation, fabric inspection, cutting, and finishing. Every clip is tied to the shift, station, and task the operator was performing.

Modality
Egocentric video + audio
Operators
Textile factory workers, multi-site
Used for
Vision-language pretraining, manipulation and defect-recognition policies
Electronics assembly
Egocentric headcam view of gloved hands soldering a motherboard on an assembly bench

8-hour egocentric days on a motherboard assembly line

Assembly line workers record full 8-hour shifts of component placement, solder inspection, and test-bench handoffs. Continuous capture preserves the transitions and idle time that short clips drop.

Modality
Egocentric video, 8 hrs/day per operator, synced task logs
Operators
Motherboard assembly line workers
Used for
Long-horizon task understanding, procedure grounding for embodied agents
Manufacturing
Egocentric view of a manufacturing plant worker inspecting metal parts at a workbench

Egocentric capture across a multi-station manufacturing plant

Plant operators wear head-mounted cameras through machine tending, changeovers, quality checks, and material handoffs across stations. Continuous capture keeps the shop-floor context that short clips lose.

Modality
Egocentric video + audio, shift-tied task logs
Operators
Manufacturing plant operators, multi-station
Used for
Manipulation policies, procedure grounding, and long-horizon task understanding for physical AI

Questions

What does Zirconoid do?

Zirconoid is a talent engine for operator data. We recruit the people who do real work and we capture what they do, then deliver those datasets to teams building physical AI. Operators join by the hour for capture scenarios, or on contract and full-time.

Who does Zirconoid work with?

Teams building physical AI — data companies, robotics organizations, and research groups. We recruit operators and run the capture. Engagements sit under a signed statement of work, license, or MSA.

What is egocentric data?

Egocentric data is video recorded from the operator's own point of view, usually with a head-mounted camera. It shows where the hands go, where the eyes go, and what the environment looks like at the moment a decision is made, which is difficult to reproduce any other way.

What kinds of datasets does Zirconoid collect?

Three programs are running now: egocentric video from textile factory floors, full 8-hour egocentric shifts on motherboard assembly lines, and egocentric capture across multi-station manufacturing plants. We take on new domains wherever humans still outperform models.

Why is human-collected data hard to replace with synthetic data?

Engineering systems that solve technical problems get commoditized. Fresh human-collected and operator-collected data does not, because it is specific to the person, the place, and the task. A real operator performs the version of a task that works on this machine, with this material, on this shift, and those deviations are the signal.

How does Zirconoid recruit and vet operators?

We source through direct relationships with employers, referrals from operators already in our network, and open recruiting in regions where an industry is concentrated. Vetting is practical for floor work: operators perform a short segment of the task on camera and a domain reviewer confirms the technique.

How are operators paid?

Operators are paid hourly for capture sessions, with a premium for full-shift recordings and for wearing equipment. Contract and full-time arrangements are available for programs that run for months.

How do I request a sample dataset?

Email [email protected] with the domain, modality, and volume you need. We return a scoped sample and a capture plan. Sample datasets are confidential and licensed for internal evaluation only.

Where does Zirconoid operate?

Worldwide. We recruit talent globally and run capture wherever the ground truth lives, from factory floors and assembly lines to manufacturing plants and workshops.

Work with us

Request a sample dataset.

Tell us the domain, modality, and volume you need. We will return a scoped sample and a capture plan.