License ~109,000 hours of manipulation data across home, manufacturing, logistics, retail and agriculture, or have data collected to your spec. Delivered in your training format.
Samples and detailed statistics available under NDA. Data can be tailored to your requirements.
Leave your contact details and requirements. We reply within 3 business days.
AI learned to think from the internet's text and images. Robots have no internet of physical work to learn from. They already have models, vision and bodies. What they lack is data from real work: diverse, task-rich data captured in homes, factories, warehouses, stores and fields.
Why RData
We don't build robots or foundation models, so your data provider is never your competitor.
~109,000 hours available today, with 20,000+ hours of monthly production capacity.
Collected on 7 commercial platforms, including bipedal humanoids, wheeled humanoids and mobile dual-arm systems.
LeRobot, RLDS, HDF5, MCAP / ROS bag, or custom.
Catalog data is held by us, so licensing terms stay clean and simple.
Data types
Combine all three for breadth, scale and grounding on real robots.
7,000 h
Operators teleoperate real robots through each task. Every episode pairs synchronized video with robot state, action trajectories and task labels. One package also includes contact-force and torque signals.
Best for: embodiment-specific fine-tuning, policy evaluation, contact-rich tasks.
80,000 h
Calibrated, synchronized three-camera capture (head + left wrist + right wrist, 1080p by default), end-effector pose and gripper-state trajectories, hand motion, task labels. Robot-agnostic and retargetable to many embodiments.
Best for: large-scale policy pre-training, cross-embodiment transfer, dexterous and deformable-object manipulation.
22,000 h
First-person video of real task processes, recorded step by step with task metadata. Mono or stereo capture.
Best for: world-model and VLA pre-training, task understanding, long-horizon planning priors.
| Modality | Teleoperation | Human Demonstration | Egocentric |
|---|---|---|---|
| Video / Image | calibrated tri-camera | mono or stereo | |
| Robot State | |||
| Motion / Trajectory | |||
| Force / Tactile | one selected package | ||
| Hand / Gripper Motion | |||
| Labels & Task Metadata |
Scenario coverage
56,000 h
Tidying, clothing, kitchenware.
View datasets →35,000 h
Fastening, connectors, precision assembly.
View datasets →6,000 h
Picking, totes, restocking.
View datasets →7,000 h
Restocking, hospitality.
View datasets →5,000 h
Picking, sorting, crating in unstructured outdoor environments.
View datasets →Delivery & quality
Delivered in your format, for example LeRobot, RLDS, HDF5, MCAP / ROS bag, or custom.
Task- and step-level labels plus task metadata. Custom schemas on request, for example language instructions, reasoning-chain annotations, or your own schema.
Tasks and scenarios are defined up front, so every hour is organized and traceable. Certified operators, trained through a multi-level certification program, collect under standardized SOPs. Automated quality gates check multimodal alignment and perform cleaning, followed by multi-stage human review. Data that fails QC is not delivered.
Need data on your own robot, in your scenario, for your task? We collect to your specifications on the same production network, with capacity of 20,000+ hours a month.
Rights & compliant delivery
Engagement models
Any package, any scenario family, or the full catalog, available now.
Collection on your embodiment, in your scenarios, for your tasks.
Ongoing dedicated supply for continuous training needs.
Tell us what you need. Data can be tailored or collected to your feedback. Pricing on request.
Send a data requestProcess
Contact
Sign NDA
Receive samples & detailed statistics
Evaluate in your pipeline
License & delivery
Who it's for
Embodiment teams
Develop skills on your hardware faster, combine real-robot data from many platforms with robot-agnostic trajectories you can retarget, and commission custom collection on your own embodiment.
Pre-training and fine-tuning
Scale pre-training and fine-tuning with diverse multimodal data organized by scenario, task and paradigm.
Academic and industry
Ready-to-use manipulation data without building a collection operation.
Resources
Plain-language guides for teams buying robot training data.
How the three main robot data paradigms differ, what each is good for, and how to combine them in a training stack.
A buyer checklist: formats, calibration and synchronization, QC, rights and consent, de-identification, and samples.
How the common robot learning data formats differ and how to pick one for your training stack.
Tell us what you need. We reply within 3 business days.