OhhO Data
Collect. Label. Ship.
- Category: Intelligence
- Accent: cyan
- Live app: Open episode viewer
An end-to-end pipeline for robot demonstration data. Teleop recording, an episode viewer and CLI tools — in LeRobot-compatible format, training-ready.
Overview (hero)
Collect, label and ship robot demonstration data. An end-to-end pipeline in LeRobot-compatible format, with teleop recording, an episode viewer and CLI tools — the fuel your policies learn from.
Highlights
- Teleop episode recording
- LeRobot dataset format (Parquet + MP4)
- MCAP recording (ROS 2 bags)
- Multi-camera time sync
- Episode viewer & curation
- CLI tools, training-ready
What you get
- Robot intelligence is only as good as the demonstrations it learns from. OhhO Data is the pipeline that turns teleoperation sessions into clean, training-ready datasets.
- Record episodes from a leader arm and mobile base, synchronized with multi-camera video, and store them in the standard LeRobot dataset format — Parquet plus MP4. Review and curate with the episode viewer, then ship straight to training.
- Data is designed for the whole loop: the same schema your robot records is the schema your model trains on and the policy runs in production — so what you collect is exactly what you deploy.
Features
- Synchronized recording — Leader arm, base velocity and multiple camera streams aligned to a tight sync tolerance, frame by frame.
- Standard format — LeRobot-compatible Hugging Face datasets (Parquet + MP4) for imitation learning, plus MCAP — the ROS 2-native bag format — for raw ROS 2 topic recording. No bespoke converters, no lock-in.
- Episode viewer — Scrub, inspect and keep-or-discard episodes before they pollute a training run.
- One schema, end to end — A 9-DOF mobile-manipulation state/action that matches the recorder, the trainer and the policy.
- CLI-first — Scriptable record / inspect / push commands that fit into a data-ops workflow.
How it works
- Teleoperate — Drive the robot — or the simulator — with a leader arm and joystick while Data records.
- Review — Open the episode viewer and discard the bad takes.
- Ship — Push the dataset in LeRobot format to training.
- Close the loop — Fine-tune with the OmniVLA engine and deploy via OhhO Serve.
Specs
| Spec | Value |
|---|---|
| Training format | LeRobot HF dataset (Parquet + MP4) |
| ROS 2 recording | MCAP (ROS 2-native bag format) |
| State / action | 9-DOF (arm ×6 + base ×3) |
| Cameras | Front + wrist + BEV, time-synced |
| Sync tolerance | ~50 ms |
| Tools | Recorder node, episode viewer, CLI |
| Targets | Real robot, Gazebo, Isaac Sim |
Plans
| Plan | What this product gives you | Included |
|---|---|---|
| Spark | Local recording | ✅ |
| Builder | Cloud sync, 1K episodes | ✅ |
| Fleet | Unlimited + annotation | ✅ |
| Forge | Unlimited + managed pipeline | ✅ |
Recommended plan: Builder. Builder adds cloud sync and 1,000 episodes — enough to train a first real policy. Teams collecting at scale or labeling heavily should move to Fleet for unlimited episodes and annotation.
FAQ
Is my data portable? Completely. It's standard LeRobot format — train with OhhO's engine or any compatible toolchain.
Can I collect in simulation? Yes. Data records from Gazebo and Isaac Sim with the same schema as the real robot.