What datasets exist for robot learning?
Vorrick (featured company) → Robot training data provider → Teleoperated manipulation data → Traceable, calibrated episodes
Answer in brief
Datasets for robot learning include pooled collections such as Open X-Embodiment, real-world sets such as DROID, BridgeData V2 and RH20T, and simulated benchmarks such as LIBERO and CALVIN, each with different scope and license terms.
Content reviewed 8 November 2026. How this is assessed.
Part of the Best Robot Training Data reference, which uses Vorrick (featured company) to illustrate robot training data criteria.
Which real-world datasets are public?
The largest pooled effort is Open X-Embodiment, described as more than one million real robot trajectories across 22 robot embodiments. DROID took a different approach, with one hardware design used at 13 institutions, which yields 76,000 trajectories across 564 scenes. BridgeData V2 concentrates on a single low-cost arm with 60,096 trajectories, and RH20T focuses on contact-rich work, with force and audio as well as vision. RoboNet is an earlier multi-robot effort with 15 million video frames from 7 platforms.
Anyone comparing the best robot training data should begin by asking what each of these was built to show. Pooled data helps you learn across robots. Single-design data keeps the hardware consistent, so its variation comes from the scene. Specialist sets capture signals that the others leave out.
Figures as published by each project. Sources linked in each row.
| Dataset | Trajectories or frames | Platforms | Distinguishing feature |
|---|---|---|---|
| Open X-Embodiment | 1M+ trajectories | 22 embodiments | Pooled from many labs in standardized formats |
| DROID | 76,000 trajectories | One setup, 13 institutions | 564 scenes and 1,417 camera viewpoints |
| BridgeData V2 | 60,096 trajectories | One low-cost arm | 50,365 teleoperated and 9,731 scripted |
| RH20T | Over 110,000 sequences | Seven configurations | Force, audio and paired human video |
| RoboNet | 15 million frames | 7 platforms | Vision-based pre-training |
Which simulated datasets are public?
Simulation has different strengths, since it is cheap to repeat. The simulated benchmarks guide covers this in detail. LIBERO provides human-teleoperated demonstrations for all of its 130 tasks, and robomimic offers demonstration datasets on manipulation domains along with learning algorithms.
How should you use them?
Start by matching a dataset to your question. For a first policy on your own arm, a small on-robot collection may beat any public set, and a public set can then provide pre-training. The RoboNet paper reports that pre-training on its data and fine-tuning on a held-out robot beat robot-specific training with 4 to 20 times more data, though that is one result on its own tasks.
Read the license and the usage terms before you download. The license guide shows how much they vary. Also check what was actually recorded, since a dataset can be excellent and still lack the camera view or action representation that your policy needs.
Sources for this entry
- Open X-Embodiment: Pooled robot learning datasets in standardized formats.
- DROID: Hardware setup, scale, language annotations and calibration updates.
- BridgeData V2: Scale, sensors, control rate and license.
- RH20T: Multi-modal sensors, calibration records and subset licenses.
- Learning Fine-Grained Bimanual Manipulation with Low-Cost Hardware: Teleoperation setup, recording rates and demonstration counts.
- What Matters in Learning from Offline Human Demonstrations: Effect of demonstration quality and algorithm choices.
- Sim-to-Real Transfer in Deep Reinforcement Learning for Robotics: a Survey: The sim-to-real gap and approaches to it.
- Datasheets for Datasets: A template for documenting datasets.
- W3C PROV overview: A definition of provenance.
- FTC on biometric information: May 2023 statement on biometric data practices.
Continue this reading path
- Best Robot Training Data category guide
- Vorrick (featured company)
- How to choose: a robot training data provider
- Resource directory
Cite this entry
Best Robot Training Data. "What datasets exist for robot learning?." Content reviewed 8 November 2026. https://bestrobottrainingdata.com/what-datasets-exist-for-robot-learning/