Robotics Data · Services · VLA Foundry
The service layer
for physical AI
PrismaX provides robot training data — teleoperation and egocentric demonstrations — and collects custom data to your spec for teams building physical AI.
Backed by
Robotics Data
The most diverse robotics data, on demand.
65 tasks · 4 arm familiesTeleoperation data
Skilled operators drive real robot arms and humanoids through real tasks. Every session is a demonstration a model can learn from directly.
Learn moreEgocentric data
First-person video of hands-on work, recorded at real work sites from the point of view of the person doing it.
Learn moreServices · Custom data collection
Data collected to your spec, delivered fast.
Tell us the tasks, robots and scenes you need. Collection starts within days on our own fleet and operator network, priced below building it yourself.
Solutions
Data for the model you’re training.
VLA training data
Language-labeled demonstrations for vision-language-action models like π0.5.
Learn more
Humanoid robot data
Bimanual arm demonstrations for the manipulation work humanoid robots are built to do.
Learn more
World model data
Egocentric and multi-view video of real physical interaction.
Learn morePricing
Subscribe today. Train today.
$1M+
And countingRecord everyday tasks.
Capture first-person video of ordinary hands-on work with a phone, action camera or headset. Every clip that meets the Operator Intent standard earns. No robot required.
For anyone with a camera
News
What we’re building, shipping, and learning.
Deep Dive
Quick Note: π0.5 Inference in 16 GB
Two fixes cut our π0.5 rollout stack from about 22 GB to 11.2 GB of VRAM, so it now fits 16 GB cards like the RTX 5080 and 5070 Ti.
Deep Dive
Latency Versus π0.5 Rollout Performance
A Python pipe buffer added hundreds of milliseconds to our π0.5 rollouts. How we found it, cut inference from 230 ms to 87 ms, and what changed on the robot.
Deep Dive
openpi / π0.5 Fine-Tuning: How-To and Tips
Ablations from a π0.5 fine-tune on YAM pick-and-place data: training scheme, lighting, inference bugs, camera placement and field of view, and backgrounds.
FAQ
Frequently asked questions
What is PrismaX?
PrismaX is the service layer for physical AI. We provide robot training data — teleoperation and egocentric demonstrations — and collect custom data to a team's spec. VLA Foundry is where contributors capture and review that data; contributors have been paid more than $1M so far.
What kinds of robot data does PrismaX offer?
Two kinds: teleoperation data, where skilled operators drive real robot arms and humanoids through tasks, and egocentric data, first-person video of hands-on work. Both are QA-scored episode by episode.
How do I buy data?
Discover and Researcher are self-serve monthly subscriptions to the data bank. For data collected to your spec, contact sales about Enterprise.
How do contributors earn with VLA Foundry?
By recording egocentric video of everyday tasks (collection is opening in waves through a waitlist), or by running teleoperation sessions on their own robots, paid per valid session.
