Egocentric
First-person capture of hands-on tasks, the scene exactly as the person performing it sees it. Point-of-view footage framed for embodied learning.
View plansThe service layerfor physical AI
Our Robotics Data gives physical-AI builders large-scale, diverse, proven demonstrations. VLA Foundry is where operators earn by capturing the real world.
Robotics Data
The most diverse robot dataset, in one console.
Teleoperation and egocentric episodes across tasks, scenes, and robot arms — searchable, filterable, downloadable.
Access DataPrismaX Consoleconsole.prismax.ai
10 datasets
flower-arrangeFranka PandaIndustrial18.4k
medicine-sortUR5Logistics9.8k
pipe-tightenxArm 7Industrial42.1k
clothes-foldALOHAOffice6.2k
fruit-sortFranka PandaKitchen12.1k
dish-rack-loadUR5Kitchen7.4k
block-stackFranka PandaIndustrial15.3k
cable-routexArm 7Industrial22.8k
parcel-scanUR5Logistics11.6k
desk-tidyALOHAOffice5.1k
Backed by
The dataGet access →
Large-scale, proven data for physical AI.
Foundation models have already trained on this data. What you get is proven in real training runs — at the scale physical AI demands.
120K+
Episodes
9,200 hrs
Duration
24x
Diversity of scenes
Teleoperation
Skilled operators teleoperate robot arms and humanoids through real tasks. Every session is a demonstration a model can learn from directly.
Plans
Subscribe today. Train today.
Self-serve at every level. Episode-based pricing. Annual saves on every plan.
Discover
For builders, hobbyists, and the AI-curious.
$9 / mo
$7 / mo billed annually
- 100 episodes / month
- 20% of the library
- Up to 640×480 @ 30fps · tasks up to 2 min
- 4 robot arm types
- Browser download
- New releases on a T+60-day window
ResearcherRecommended
For academic researchers and ML engineers exploring physical AI.
$89 / mo
$71 / mo billed annually
- 1,500 episodes / month
- 60% of the library
- Up to 1080p @ 60fps · tasks up to 3 min
- 8 robot arm types
- Browser download + API access
- New releases on a T+30-day window
Enterprise
For startups and companies training physical AI models.
$899 / mo
$719 / mo billed annually
- 18,000 episodes / month
- 100% of the library — full raw access
- Full resolution · all task lengths
- 12+ robot arm types — all supported
- Browser + API + JSON download
- New releases day one (T+0)
- Self-serve — no sales contract
The network
Earn by teaching robots.
Record everyday tasks.
Capture first-person video of ordinary hands-on work with a headset or camera. Every clip that meets the Operator Intent standard earns. No robot required.
For anyone with a cameraStart contributing →Total paid to contributors
$1M+
and counting
Put your robots to work.
Run teleoperation sessions on your own hardware, on real tasks backed by real demand. Paid per valid session, remote and on your own schedule.
For owner-operators with robotsApply as an operator →Field notes
See all writing →What we’re learning from real operation.
What smoothness really measures in operator data.
Inside the Operator Intent standard — why smoothness, not just task success, is the signal that separates trainable demonstrations from noise.
Read →
Sourcing tasks from real demand, not synthetic prompts.
How VLA Foundry decides what to capture next — and why every task traces back to a real robotics need rather than a generated wishlist.
Read →
The feedback loop between training and QA criteria.
Lessons from training internal models on Foundry output keep rewriting our review criteria. A look at the loop that keeps the data improving.
Read →Upcoming
- Sep 2026TechCrunch Disrupt San FranciscoRegister →
- Oct 2026IROS Vienna, AustriaRegister →
Get started
Train on the data, or get paid to capture it.
Same QA-scored data, same operator network, at every tier — only the scale changes. Every episode was captured by someone getting paid to do it.