A-01
First-person capture
Head- and chest-mounted egocentric video of real tasks in real environments: homes, kitchens, workshops and retail. The viewpoint your model will actually see.
- Viewpoint
- Egocentric
- Setting
- Real environments
FIG. 01 / FIELD DATA FOR AI
Measured in the real world. Delivered to your model.
Spatial Theory sends trained people into real places to capture egocentric video and spatial data, then verifies every session by hand before it reaches you.
- Egocentric video
- Multi-sensor capture
- Geo-tagged sessions
- Human-verified QA
- Project
- Spatial Theory
- Drawing
- ST-001
- Revision
- A
- Scale
- 1:1 real world
01WHAT WE DO
Models can't learn the physical world from the internet alone.
Robots and multimodal models need to see how people actually move through and act on the world. We produce that data with the consistency and paper trail a research team can rely on.
A-02
Spatial and sensor data
Every session is anchored to a place with geohash and venue metadata, so you know where data came from and can balance coverage across cities and environments.
- Anchor
- Venue and geohash
- Coverage
- Balanced by city
A-03
Verified by humans
Nothing ships unreviewed. Each session passes through a structured QA pipeline with accept, reject and re-capture loops, so what you receive matches the spec.
- Review
- Every session
- Outcome
- Accept, reject, re-capture
BUILT FOR
- Robotics
- Embodied AI
- World models
- Spatial understanding
- AR and VR
02HOW IT WORKS
From brief to delivered dataset in four steps.
- 01
Scope
You tell us the tasks, environments, volume and format. We turn it into a capture spec and a deployment plan.
- 02
Deploy
We stand up sessions at venues in the cities you need, with trained contributors and on-site leads.
- 03
Capture
Contributors record the target tasks using standardised guidelines, equipment and session metadata.
- 04
Verify and deliver
Reviewers QA every session. Approved data is packaged and delivered in your format, with provenance attached.
SESSION OVERVIEWILLUSTRATIVE
03QUALITY
Quality is the product, not an afterthought.
Bad data is expensive. It's a failed training run you only discover weeks later. We catch problems at the source, before they reach you.
- Standardised capture guidelines for every task type
- Required metadata on every session: venue, location, contributor skill group
- Human review of every submission, with re-capture when it misses spec
- Full traceability from delivered clip back to session and site
04TRUST AND PRIVACY
Built to be handed to a legal team.
T-01
Consent first
Contributors and venues agree to clear terms before anything is recorded. No hidden capture.
T-02
Privacy by design
Capture guidelines keep bystanders and sensitive information out of frame, and flagged footage is removed in review.
T-03
Clear provenance
Every clip carries origin, consent and QA records, so you can show where your data came from.
Want to contribute or host a session?
We work with individuals and venues. Tell us where you are and we'll be in touch about upcoming deployments.
05FAQ
Common questions
What kind of data do you collect?
Primarily egocentric (first-person) video of everyday and task-specific activity, with session-level spatial and venue metadata. Custom capture specs are available.
Who is this for?
AI labs, robotics companies and research teams that need real-world data they can't get from scraped or synthetic sources.
How do you make sure the data is good?
Every session is reviewed by a human against the capture spec. Sessions that miss spec are rejected or re-captured before delivery.
Where do you operate?
We run deployments in multiple cities through on-site leads and partner venues. Get in touch with your target regions and we'll confirm coverage.
How do I work with you?
Email us with your use case and volume. We'll come back with a proposed spec, timeline and pilot.
06CONTACT
Let's capture what your model is missing.
Tell us what you're building. We'll reply within one business day.
hello@spatialtheory.org