Teleoperation data
Teleoperation data for robot learning.
Teleoperation data is a record of a skilled human operator driving a real robot through a task: synchronized video from the robot’s overhead and wrist cameras, plus the joint states needed to reconstruct every movement.
PrismaX collects it from two sources, our own fleet of 33 YAM dual-arm rigs and VLA Foundry operators’ robots, and scores every episode before it ships.
- 65
- Tasks in the catalog
- 7
- Scenes
- 4
- Arm families
- 33
- YAM dual-arm rigs in our own fleet
Specification
What every episode contains.
| Embodiments | Bimanual serial arms with 6 or 7 degrees of freedom, with 1-DoF grippers or dexterous hands |
|---|---|
| Robot arms | RLM 63, YAM, Piper, TOK2 in the catalog; 12+ arm types supported on Enterprise |
| Cameras | An overhead camera with the full workspace in view, plus at least one camera on each wrist |
| Robot state | Angles of every joint; rigs in our own fleet each have an individually measured URDF |
| Resolution, frame rate, length |
|
| Delivery | Discover: Browser download · Researcher: Browser download + API access · Enterprise: Browser + API + JSON download |
| Episode file format | TBD |
| Annotations | Task prompt and validator quality score on every episode; further labels TBD |
Example tasks
Straight from the catalog.
| Task | Robot arms | Scene | Episodes |
|---|---|---|---|
| Biology Lab | YAM | Lab and Research | 108.8k |
| Pick and place various fruits into containers | Piper · RLM 63 · TOK2 · YAM | Kitchen | 2.8k |
| Pick and place flowers into vase and vase on shelf | Piper · RLM 63 · TOK2 · YAM | Home | 2.0k |
| Block manipulation | YAM | Home | 2.0k |
| Pack box of components | YAM | Logistics | 1.8k |
| Pick and place slippers into bin | Piper · RLM 63 · TOK2 · YAM | Home | 1.6k |
| Tying shoelaces | YAM | Home | 1.5k |
| Clean up kitchen area | YAM | Kitchen | 1.4k |
| Unpack box of objects into bins | YAM | Logistics | 1.2k |
| Pick and place school supplies | Piper · TOK2 · YAM | Office | 1.1k |
Operators drive robot arms through six everyday tasks: folding a T-shirt, sorting fruit, racking dishes, peeling a banana, tidying desk items and a shape sorter.
Two robot arms arrange flowers in a vase, seen from the overhead and wrist cameras at the same time.
A montage of teleoperated sessions folding T-shirts, seen from many angles.
How we collect
Two sources, one standard.
33 YAM dual-arm rigs
RLM 63, Piper and TOK2
Standards that evolve
Quality and delivery
Software checks camera coverage, joint states, image quality and sample rate on every episode. Validators then judge each demonstration pass or fail and score its speed, smoothness and completion. Episodes ship by browser download, API or JSON export, depending on your plan.
Related reading
All news
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.
Perspective
Standards for Data
The first arc of the PrismaX Standards for Data: repeatable guidelines for collecting pretraining-grade robotics data, and the reference our validators use to apply them.
Deep Dive
Teleoperation: Now and the Future of AI Robotics
Read this article if you’re interested in contributing to the robotics revolution, which will reshape the way we work and interact with the physical world.
FAQ
Frequently asked questions
What is teleoperation data?
Teleoperation data is a record of a human operator driving a real robot through a task: synchronized video from the robot's overhead and wrist cameras plus the joint angles needed to reconstruct every movement. Models learn to map what the cameras see to the actions the operator took.
Which robots is PrismaX teleoperation data recorded on?
The catalog covers RLM 63, YAM, Piper, TOK2 arms. Episodes come from two sources: our own fleet of 33 YAM dual-arm rigs, and VLA Foundry operators running sessions on their own robots, including RLM 63, Piper and TOK2 arms. Enterprise supports 12+ arm types.
Which cameras and sensors are included?
At minimum an overhead camera with the whole workspace in view and one camera on each wrist, plus the angle of every joint. Rigs in our own fleet also ship with an individually measured URDF and globally available camera SKUs, so the setup can be reproduced.
How is teleoperation data different from plain video?
Video shows what happened; teleoperation data also records what the robot did. Because joint states come with every frame, a model can learn actions directly instead of inferring them from pixels.
Can PrismaX collect teleoperation data for my tasks or robot?
Yes. Custom data collection, part of Enterprise, records the tasks, robots, scenes and volume you specify, held to the same quality standard as the catalog. Collection starts within days.
Next step
Teleoperation data for your tasks.
Start from the catalog, or tell us the tasks, robots and scenes you need collected.