Data Collection & Annotation
for Physical Intelligence
We crowdsource video collectors for robotics training from actual day to day operations and/or controlled environments. Annotated datasets, ready to use.
actual organic recordings
controlled environments
1080p · 30/60fps
01 · INTRO
Video data for AI and robotics.
We run people management across multiple businesses, recording egocentric video with head-mounted GoPros and smartphones across hotels, warehouses, kitchens, and day-to-day processes. We film and provide the raw material — or fully annotate it with the required internal labelling.
02 · FACILITY
A dedicated facility. Custom scenarios.
Conscious of the importance of variability (lighting, people, etc.), we build the environment your model needs to see. Functional kitchens, hotel rooms, restaurant stations, warehouse aisles and shelving — all set up and controlled to capture consistent, reproducible data.
- Kitchen tasks
- Hotel operations
- Restaurant workflows
- Warehouse activities
03 · SPECS
Egocentric. 1080p. 30 or 60 fps.
Head-mounted cameras capture the exact first-person perspective an embodied agent needs to learn. Consistent image quality, precise sync, and structured metadata on every clip.
04 · RECORDING TYPES
We cover every angle of manipulation.
Every project is set up with the capture modalities your pipeline requires.
- Egocentric (first person, head mounted)
- Hand manipulation and object interaction
- Real task execution
- Full body movement
- Multi-angle capture
- Spatial kinematic data (6DoF pose, trajectories)
05 · CONSENT & LEGAL
Every recording comes with signed consent.
All participants sign consent forms before every recording, so your data is fully cleared for commercial AI training use. No legal ambiguity, no redistribution risk.
06 · SCALE & FAILURE DATA
Thousands of recordings. Failures included.
We capture thousands of task-specific egocentric recordings across multiple environments. Every project includes both successful task executions and the failures — because that's where models learn to recover.
“We capture failure modes properly. Most vendors throw them away.”
07 · LANGUAGE
English VoiceOver option.
If the language in the video is other than English (e.g. Spanish, Greek), we can accommodate English as VoiceOver depending on project needs.
08 · DELIVERY
How the data is delivered.
We deliver data through the buckets you already use in your pipeline, with annotations in COCO JSON format ready for training.
- AWS S3 buckets
- Google Cloud Storage buckets
- Annotations delivered in COCO JSON format
> Video Session initialized. Step 1 of 2.
Egocentric Video Collection
Hand manipulating a pruner to cut leaves.
09 · FAQ
FAQs
Either fill out the form, or send us an email at projects@nosisdata.com
We first conduct an intro call to understand the scope, define the golden dataset, understand the complexity of the crowdsourcing efforts and equipment required. We then define whether the end goal is completed tasks or hours of specific content, and we then proceed to share a transparent quote.
We have dedicated cameras with 6 axis IMU sensors and ultrawide lenses, specialized for data collection. We also use suitable phones with the required sensors (iPhones 11+, Samsung Galaxys, Xiaomi Redmis, etc.). Email us for the specifics. We can also accomodate specific requests and/or add more sensors depending on the project needs.
We have partnerships with business in LATAM, and Greece that facilitates recordings. Nonetheless, we also have custom recruit new partnerships depending on the projects.
Yes. Share the specifications, number of completed tasks / hours and we proceed accordingly.
Yes. We can accommodate annotation services as well. Though that's treated differently.
Tell us what you're training.
Tell us precisely what you want, and we'll deliver a first quality check sample.
> Intake session initialized.
Step 1 of 7Or email us directly at info@nosisdata.com