Real-world video data for frontier AI labs.

We film skilled people doing real jobs, annotate the footage to your spec, and deliver quality-checked datasets ready to train on.

10,000
Humans available to collect
25,000
Hours collected per day
5
Countries in active collection
96+
Tasks across 18 sectors

HiOps is a video data provider for physical AI. We film skilled workers doing real jobs with first-person and multi-view cameras, annotate the footage to a lab's specification, and deliver quality-checked datasets into the lab's own storage. Based in Los Angeles with operations in Kathmandu, collecting in five countries.

Built with the platforms frontier labs run on.

NVIDIA Inception Program
Amazon Web Services
Google Cloud

HiOps is a member of the NVIDIA Inception program, and our collection, annotation, and delivery pipelines run on AWS and Google Cloud. Learn more

Video data collection, made ready the way you want it.

Egocentric video

First-person footage from head-mounted cameras worn by skilled workers doing their actual jobs. Hands, tools, and objects in frame, at the viewpoint a manipulation model learns from.

Multi-view video

Fixed cameras around the workstation capture the same task from the outside, synchronized with the egocentric stream for pose, reconstruction, and action understanding.

Annotation to your spec

Task and sub-task segmentation, language descriptions, success and failure labels, hand and object tracking. Delivered in the schema you train on.

96+ tasks across 18 sectors.

Every task below is in active collection. Search for the one your model is failing on and request a sample.

96 / 96 tasks · 18 sectors

How it works

  1. 01

    Define your data requirements

    Tell us the tasks, environments, viewpoints, episode lengths, and labels your model needs. We write it into a collection spec and produce a sample batch before scaling.

  2. 02

    We collect and annotate

    Our operators and partner facilities capture footage in real workplaces across five countries. Every episode is annotated by trained reviewers against the spec.

  3. 03

    Deliver QC-verified datasets

    Multi-layer quality control filters bad takes before delivery. Batches land in your storage with manifests and acceptance rates, ready to train on.

Built for teams training the frontier.

Headquartered in Los Angeles, California, with an operations office in Kathmandu, Nepal, and collection running in five countries.

Real workplaces, not studios
Farms, kitchens, warehouses, workshops. The lighting, clutter, and variation that staged collection never produces.
One team, start to finish
Collection, annotation, quality control, and delivery are one accountable team.
Spec-driven
Each engagement starts with what your model fails on, not with a fixed dataset.
Consented and cleared
Every participant is consented and compensated. Rights are cleared for commercial model training.

Common questions

What does HiOps provide?
HiOps provides real-world video training data for physical AI and robotics. We film skilled workers doing real jobs with egocentric (head-mounted) and multi-view cameras, annotate the footage to your specification, and deliver quality-checked datasets in your format.
What kinds of tasks can HiOps collect?
HiOps collects 96+ task types across 18 sectors, including agriculture, construction, manufacturing, logistics, commercial kitchens, hospitality, retail, auto services, and more. Custom tasks can be scoped on request.
What data formats and annotations are available?
Egocentric video, synchronized multi-view video, task and sub-task segmentation, natural-language descriptions, success and failure labels, and hand and object tracking. Delivery is in the schema your training pipeline expects, into your own AWS S3 or Google Cloud Storage bucket.
How do I evaluate the data before committing?
Request a free sample batch. We reply within 15 minutes during business hours and send sample data with annotations within a business day. A scoped pilot program is available for larger evaluations.
Are participants consented and is the data cleared for commercial use?
Yes. Every person who appears in HiOps footage takes part under written consent and is compensated. Rights are cleared for commercial model training before delivery.
Where is HiOps based?
HiOps is headquartered in Los Angeles, California, with an operations office in Kathmandu, Nepal, and data collection running in five countries.

See the data before you commit to it.

Tell us the sector and task. We send a sample batch with annotations and a collection plan, usually within a business day.