Open robot dataset spec ledger
Answer in brief
Of 25 open academic robot datasets, 15 state a control rate, 8 state terms for the data, 9 state recording hours and 1 publishes a dated calibration release. Every value on this page links to the page it came from.
Content reviewed 8 November 2026. How this is assessed.
Vorrick and this reference are run by the same operator. Vorrick is excluded from this page, and this page ranks nothing.
What is in this dataset spec ledger?
This ledger records 25 open academic robot datasets across 13 specification columns, sorted alphabetically, with 152 sourced values and 151 cells marked as not stated. The data is free to download as ledger.csv.
Each value links to the page it came from and shows the date it was read. Numbers were confirmed as exact strings in the text of the cited page. Where a page did not give a value, the cell says so. Nothing is inferred.
Selection rule: a dataset is listed if a university or public research institute is a releasing institution. A dataset released by a commercial company alone is excluded. Some listed entries are collaborations that include company staff or use company storage, and they stay because academic institutions are co-releasers.
Vendors of hardware are described by type, for example 7-DoF arm or stereo RGB camera, so that this page names no commercial product.
How large are open robot datasets?
21 of 25 datasets state a trajectory, episode or demonstration count and 9 state hours; counts use different units and are shown as each project reports them, so they are not comparable line by line (for example DROID).
| Dataset | Trajectories or episodes | Hours | Robot type |
|---|---|---|---|
| ARIO | approximately 3 million episodes source read 8 Nov 2026 | Not stated on the project page | Mixed robot series, 258 series named source read 8 Nov 2026 |
| BiPlay | 7023 clips source read 8 Nov 2026 | 10 source read 8 Nov 2026 | Bimanual teleoperated setup source read 8 Nov 2026 |
| BridgeData V1 | 7,200 demonstrations source read 8 Nov 2026 | Not stated on the project page | 6-DoF low-cost arm source read 8 Nov 2026 |
| BridgeData V2 | 60,096 trajectories source read 8 Nov 2026 | Not stated on the project page | Low-cost arm source read 8 Nov 2026 |
| CALVIN | Not stated on the project page | 24 hours of teleoperated play source read 8 Nov 2026 | Simulated 7-DoF arm source read 8 Nov 2026 |
| D3IL | Not stated on the project page | Not stated on the project page | Simulated 7-DoF arm source read 8 Nov 2026 |
| DROID | 76k demonstration trajectories source read 8 Nov 2026 | 350 source read 8 Nov 2026 | 7-DoF arm, one shared setup at 13 institutions source read 8 Nov 2026 |
| FMB | 22,550 trajectories source read 8 Nov 2026 | Not stated on the project page | Single arm source read 8 Nov 2026 |
| FurnitureBench | 5100 teleoperation demonstrations source read 8 Nov 2026 | 219.6 source read 8 Nov 2026 | 7-DoF arm source read 8 Nov 2026 |
| Homes of New York | 5620 demonstrations source read 8 Nov 2026 | 13 source read 8 Nov 2026 | Hand-held demonstration tool source read 8 Nov 2026 |
| Humanoid Everyday | 10.3k trajectories source read 8 Nov 2026 | Not stated on the project page | Humanoid source read 8 Nov 2026 |
| LIBERO | 50 demonstrations per task source read 8 Nov 2026 | Not stated on the project page | Not stated on the project page |
| ManiSkill2 | Not stated on the project page | Not stated on the project page | Not stated on the project page |
| MIME | 8260 demonstrations source read 8 Nov 2026 | Not stated on the project page | Two-arm robot, 7-DoF arms source read 8 Nov 2026 |
| Open X-Embodiment | 1M+ real robot trajectories source read 8 Nov 2026 | Not stated on the project page | 22 robot embodiments, pooled source read 8 Nov 2026 |
| Real Robot Challenge datasets | Not stated on the project page | Not stated on the project page | Three-finger open-design platform source read 8 Nov 2026 |
| RH20T | Over 110,000 sequences source read 8 Nov 2026 | Not stated on the project page | Arm with force-torque sensor, configurations 1 to 7 source read 8 Nov 2026 |
| RLBench | Not limited: demonstrations generated by motion planners source read 8 Nov 2026 | Not stated on the project page | Simulated arm source read 8 Nov 2026 |
| RoboCasa | 2024 paper: 1,250 human demonstrations; 72,000 machine-generated source read 8 Nov 2026 | 2026 release: 600+ hours human, 1,600+ hours machine-generated source read 8 Nov 2026 | Simulated arm on a mobile base source read 8 Nov 2026 |
| robomimic | 200 per proficient-human dataset source read 8 Nov 2026 | Not stated on the project page | Arm, simulation and real hardware source read 8 Nov 2026 |
| RoboNet | Roughly 162,000 trajectories source read 8 Nov 2026 | Not stated on the project page | 7 robot platforms source read 8 Nov 2026 |
| RoboSet | 100,050 trajectories in the full set; 7,500 in the kitchen subset source read 8 Nov 2026 | Not stated on the project page | Commodity arm source read 8 Nov 2026 |
| RoboTurk pilot (simulated) | Over 2200 demonstrations source read 8 Nov 2026 | 137.5 on the project page; 137 in the paper source 1 source 2 read 8 Nov 2026 | Simulated 7-DoF arm source read 8 Nov 2026 |
| RoboTurk real-world | 2144 demonstrations source read 8 Nov 2026 | 111 source read 8 Nov 2026 | Not stated on the project page |
| Universal Manipulation Interface | 1400 demonstrations, cup arrangement set source read 8 Nov 2026 | 12 person-hours, cup arrangement set source read 8 Nov 2026 | Hand-held gripper for collection source read 8 Nov 2026 |
What control rate do open robot datasets use?
15 of 25 datasets state a control or recording rate, from 5 Hz (BridgeData V2) to 200 Hz (ARIO); the highest figure is a proprioception stream, so read the cell, not just the number.
| Dataset | Control rate | Camera count and placement |
|---|---|---|
| ARIO | Proprioception 200 Hz, on the project's own platform source read 8 Nov 2026 | 3 RGB-D, on the project's own platform source read 8 Nov 2026 |
| BiPlay | Not stated on the project page | Not stated on the project page |
| BridgeData V1 | Not stated on the project page | 3 to 5; one fixed, others on flexible mounts source read 8 Nov 2026 |
| BridgeData V2 | 5 Hz source read 8 Nov 2026 | 4: one fixed over-the-shoulder RGB-D, two randomized RGB, one wrist RGB source read 8 Nov 2026 |
| CALVIN | 30 Hz source read 8 Nov 2026 | 2: one fixed, one on the gripper source read 8 Nov 2026 |
| D3IL | Not stated on the project page | 2: in-hand and front view source read 8 Nov 2026 |
| DROID | 15 Hz source read 8 Nov 2026 | 3: two adjustable stereo, one wrist-mounted stereo source read 8 Nov 2026 |
| FMB | 10 Hz teleoperation command source read 8 Nov 2026 | 4: two wrist, two attached to the environment source read 8 Nov 2026 |
| FurnitureBench | 10 Hz commands source read 8 Nov 2026 | 3 RGB-D cameras; front and wrist views are the observation inputs source read 8 Nov 2026 |
| Homes of New York | 30 Hz recording source read 8 Nov 2026 | Phone camera on the hand-held tool, count not stated source read 8 Nov 2026 |
| Humanoid Everyday | 30 Hz capture source read 8 Nov 2026 | RGB-D cameras and LiDAR, count not stated source read 8 Nov 2026 |
| LIBERO | Not stated on the project page | Not stated on the project page |
| ManiSkill2 | 20 Hz, simulation source read 8 Nov 2026 | 2 in the baseline data: base camera and hand camera source read 8 Nov 2026 |
| MIME | Not stated on the project page | 3: head depth camera, one on each wrist source read 8 Nov 2026 |
| Open X-Embodiment | Not stated on the project page | Varies by source dataset source read 8 Nov 2026 |
| Real Robot Challenge datasets | 50 Hz source read 8 Nov 2026 | 3 global-shutter cameras source read 8 Nov 2026 |
| RH20T | Joints 10 Hz, gripper pose 100 Hz source read 8 Nov 2026 | 9 to 12: 1-2 in-hand, 8-10 global RGB-D source read 8 Nov 2026 |
| RLBench | Not stated on the project page | Over-the-shoulder and eye-in-hand source read 8 Nov 2026 |
| RoboCasa | 20 Hz source read 8 Nov 2026 | 3: one eye-in-hand, two workspace cameras source read 8 Nov 2026 |
| robomimic | 20 Hz policy actions source read 8 Nov 2026 | External camera and wrist camera per arm source read 8 Nov 2026 |
| RoboNet | Not stated on the project page | Over a hundred viewpoints, count per robot not stated source read 8 Nov 2026 |
| RoboSet | 5 Hz source read 8 Nov 2026 | 4: three fixed (top, left, right), one wrist source read 8 Nov 2026 |
| RoboTurk pilot (simulated) | Not stated on the project page | Not stated on the project page |
| RoboTurk real-world | Not stated on the project page | Not stated on the project page |
| Universal Manipulation Interface | Observations resampled, often 10-20 Hz source read 8 Nov 2026 | Wrist-mounted fisheye camera source read 8 Nov 2026 |
How many open robot datasets record depth, force-torque or tactile data?
16 of 25 datasets state some depth data, 3 state force or torque data and 5 state tactile data (RH20T states all three).
| Dataset | Depth | Force-torque | Tactile |
|---|---|---|---|
| ARIO | Yes, on the project's own platform source read 8 Nov 2026 | Not stated on the project page | Yes, tactile aligned at 100 Hz source read 8 Nov 2026 |
| BiPlay | Not stated on the project page | Not stated on the project page | Not stated on the project page |
| BridgeData V1 | Partly, depth cameras in part of the data source read 8 Nov 2026 | Not stated on the project page | Not stated on the project page |
| BridgeData V2 | Partly, portions of the data source read 8 Nov 2026 | Not stated on the project page | Not stated on the project page |
| CALVIN | Yes source read 8 Nov 2026 | Not stated on the project page | Yes, simulated tactile image source read 8 Nov 2026 |
| D3IL | Not stated on the project page | Not stated on the project page | Not stated on the project page |
| DROID | Yes source read 8 Nov 2026 | Not stated on the project page | Not stated on the project page |
| FMB | Yes source read 8 Nov 2026 | Yes, end-effector force and torque source read 8 Nov 2026 | Not stated on the project page |
| FurnitureBench | Yes, front and wrist depth source read 8 Nov 2026 | Not stated on the project page | Not stated on the project page |
| Homes of New York | Yes, 1.5 million RGB-D frames source read 8 Nov 2026 | No: the paper says the tool lacks force sensors source read 8 Nov 2026 | No: the paper says the tool lacks tactile sensors source read 8 Nov 2026 |
| Humanoid Everyday | Yes source read 8 Nov 2026 | Not stated on the project page | Yes source read 8 Nov 2026 |
| LIBERO | Not stated on the project page | Not stated on the project page | Not stated on the project page |
| ManiSkill2 | Yes source read 8 Nov 2026 | Not stated on the project page | Not stated on the project page |
| MIME | Yes source read 8 Nov 2026 | Not stated on the project page | Not stated on the project page |
| Open X-Embodiment | Varies by source dataset source read 8 Nov 2026 | Not stated on the project page | Not stated on the project page |
| Real Robot Challenge datasets | Not stated on the project page | Not stated on the project page | Basic, from sensors in the fingertips source read 8 Nov 2026 |
| RH20T | Yes, 10 Hz source read 8 Nov 2026 | Yes, 6-DoF at 100 Hz source read 8 Nov 2026 | Yes, configuration 7 only, 200 Hz source read 8 Nov 2026 |
| RLBench | Yes source read 8 Nov 2026 | Joint forces only source read 8 Nov 2026 | Not stated on the project page |
| RoboCasa | Not stated on the project page | Not stated on the project page | Not stated on the project page |
| robomimic | Not stated on the project page | Not stated on the project page | Not stated on the project page |
| RoboNet | Not stated on the project page | Not stated on the project page | Not stated on the project page |
| RoboSet | Yes source read 8 Nov 2026 | Not stated on the project page | Not stated on the project page |
| RoboTurk pilot (simulated) | Not stated on the project page | Not stated on the project page | Not stated on the project page |
| RoboTurk real-world | Yes, depth videos source read 8 Nov 2026 | Not stated on the project page | Not stated on the project page |
| Universal Manipulation Interface | Not stated on the project page | Not stated on the project page | Not stated on the project page |
Which open robot datasets have language annotations?
7 of 25 datasets state language annotation, with coverage from 1 percent of the data (CALVIN) to all of it (BridgeData V2).
| Dataset | Language annotation |
|---|---|
| ARIO | Not stated on the project page |
| BiPlay | Yes, language annotated source read 8 Nov 2026 |
| BridgeData V1 | Not stated on the project page |
| BridgeData V2 | Yes, all data source read 8 Nov 2026 |
| CALVIN | Yes, 1 percent of the data annotated source read 8 Nov 2026 |
| D3IL | Not stated on the project page |
| DROID | Yes: 3 annotations for 95 percent of successful episodes (75k), December 2024 source read 8 Nov 2026 |
| FMB | Not stated on the project page |
| FurnitureBench | Not stated on the project page |
| Homes of New York | Not stated on the project page |
| Humanoid Everyday | Yes source read 8 Nov 2026 |
| LIBERO | Yes, language-conditioned tasks source read 8 Nov 2026 |
| ManiSkill2 | Not stated on the project page |
| MIME | Not stated on the project page |
| Open X-Embodiment | Not stated on the project page |
| Real Robot Challenge datasets | Not stated on the project page |
| RH20T | Not stated on the project page |
| RLBench | Not stated on the project page |
| RoboCasa | Not stated on the project page |
| robomimic | Not stated on the project page |
| RoboNet | Not stated on the project page |
| RoboSet | Yes, language instructions source read 8 Nov 2026 |
| RoboTurk pilot (simulated) | Not stated on the project page |
| RoboTurk real-world | Not stated on the project page |
| Universal Manipulation Interface | Not stated on the project page |
Which robot datasets publish a calibration release date?
1 of 25 datasets give a dated calibration release: April 2025, improved calibrations for 36k episodes (DROID). Calibration files without a date count as not stated.
| Dataset | Calibration release date |
|---|---|
| ARIO | Not stated on the project page |
| BiPlay | Not stated on the project page |
| BridgeData V1 | Not stated on the project page |
| BridgeData V2 | Not stated on the project page |
| CALVIN | Not stated on the project page |
| D3IL | Not stated on the project page |
| DROID | April 2025, improved calibrations for 36k episodes source read 8 Nov 2026 |
| FMB | Not stated on the project page |
| FurnitureBench | Not stated on the project page |
| Homes of New York | Not stated on the project page |
| Humanoid Everyday | Not stated on the project page |
| LIBERO | Not stated on the project page |
| ManiSkill2 | Not stated on the project page |
| MIME | Not stated on the project page |
| Open X-Embodiment | Not stated on the project page |
| Real Robot Challenge datasets | Not stated on the project page |
| RH20T | Not stated on the project page |
| RLBench | Not stated on the project page |
| RoboCasa | Not stated on the project page |
| robomimic | Not stated on the project page |
| RoboNet | Not stated on the project page |
| RoboSet | Not stated on the project page |
| RoboTurk pilot (simulated) | Not stated on the project page |
| RoboTurk real-world | Not stated on the project page |
| Universal Manipulation Interface | Not stated on the project page |
What formats and licences do open robot datasets use?
8 of 25 datasets state terms for the data, of which 7 are CC BY 4.0 (BridgeData V2); 9 state terms for code only and 8 state none. A code licence is not a data licence, and this page keeps them apart.
10 of 25 datasets state a file format. This is not legal advice. Read the terms at the source before you use a dataset.
| Dataset | Format | Licence |
|---|---|---|
| ARIO | Not stated on the project page | Not stated on the project page |
| BiPlay | Not stated on the project page | Not stated on the project page |
| BridgeData V1 | Not stated on the project page | Not stated on the project page |
| BridgeData V2 | JPEG images source read 8 Nov 2026 | CC BY 4.0 source read 8 Nov 2026 |
| CALVIN | Not stated on the project page | Code only: MIT. Data terms not stated source read 8 Nov 2026 |
| D3IL | Not stated on the project page | Code only: MIT. Data terms not stated source read 8 Nov 2026 |
| DROID | RLDS, plus raw MP4, H5 and SVO files source read 8 Nov 2026 | CC BY 4.0 source read 8 Nov 2026 |
| FMB | Not stated on the project page | CC BY 4.0 source read 8 Nov 2026 |
| FurnitureBench | Not stated on the project page | Code only: MIT. Data terms not stated source read 8 Nov 2026 |
| Homes of New York | Not stated on the project page | Code only: MIT. Paper says data is under a permissive license, not named source read 8 Nov 2026 |
| Humanoid Everyday | Not stated on the project page | Not stated on the project page |
| LIBERO | Not stated on the project page | CC BY 4.0 on the repository page. Code: MIT source read 8 Nov 2026 |
| ManiSkill2 | Not stated on the project page | Code only: Apache 2.0. Assets: CC BY-NC 4.0 source read 8 Nov 2026 |
| MIME | Not stated on the project page | Not stated on the project page |
| Open X-Embodiment | RLDS source read 8 Nov 2026 | Software Apache 2.0, other materials CC BY 4.0 source read 8 Nov 2026 |
| Real Robot Challenge datasets | HDF5 in the paper; zarr by default in the repository source 1 source 2 read 8 Nov 2026 | Code only: BSD-3-Clause. Data terms not stated source read 8 Nov 2026 |
| RH20T | MP4 video, NPY arrays, JSON metadata source read 8 Nov 2026 | Subsets: CC BY-SA 4.0 and CC BY-NC 4.0 source read 8 Nov 2026 |
| RLBench | Not stated on the project page | Code only: non-commercial research licence. Data terms not stated source read 8 Nov 2026 |
| RoboCasa | Not stated on the project page | Assets and datasets: CC BY 4.0 source read 8 Nov 2026 |
| robomimic | HDF5 source read 8 Nov 2026 | Code only: MIT. Data terms not stated source read 8 Nov 2026 |
| RoboNet | HDF5, videos as MP4 source read 8 Nov 2026 | Dataset licence file: CC BY 4.0. Code: MIT source read 8 Nov 2026 |
| RoboSet | HDF5 source read 8 Nov 2026 | Not stated on the project page |
| RoboTurk pilot (simulated) | Not stated on the project page | Not stated on the project page |
| RoboTurk real-world | MP4 video source read 8 Nov 2026 | Not stated on the project page |
| Universal Manipulation Interface | Zarr source read 8 Nov 2026 | Code only: MIT. Data terms not stated source read 8 Nov 2026 |
Where do primary sources disagree about these datasets?
3 of 25 datasets have a disagreement between two primary pages read for this ledger (Real Robot Challenge datasets), and both versions are shown.
| Dataset | Sources disagree |
|---|---|
| ARIO | None found |
| BiPlay | None found |
| BridgeData V1 | None found |
| BridgeData V2 | None found |
| CALVIN | None found |
| D3IL | None found |
| DROID | None found |
| FMB | None found |
| FurnitureBench | None found |
| Homes of New York | None found |
| Humanoid Everyday | None found |
| LIBERO | None found |
| ManiSkill2 | None found |
| MIME | None found |
| Open X-Embodiment | None found |
| Real Robot Challenge datasets | Paper: a single HDF5 file. Repository: zarr files by default source 1 source 2 read 8 Nov 2026 |
| RH20T | None found |
| RLBench | None found |
| RoboCasa | Paper reports 1,250 human demonstrations; 2026 release page reports 600+ hours. Figures describe different releases source 1 source 2 read 8 Nov 2026 |
| robomimic | None found |
| RoboNet | None found |
| RoboSet | None found |
| RoboTurk pilot (simulated) | Project page: 137.5 hrs. Paper: 137 hours source 1 source 2 read 8 Nov 2026 |
| RoboTurk real-world | None found |
| Universal Manipulation Interface | None found |
How was this ledger checked?
Pages were read in full as text, using an ordinary request with no altered headers, and every number was searched for as an exact string on the cited page. A page that refused was recorded and not worked around.
"Not stated on the project page" means the value was not found on the project page, paper or repository read for that dataset on 8 Nov 2026. It does not mean the value is unknown to the project.
Results from this page describe what projects publish about their own data. They say nothing about data quality, and this page does not compare or recommend datasets.
Sources for this entry
- ARIO paper: Read as text on 8 Nov 2026.
- BiPlay paper: Read as text on 8 Nov 2026.
- BridgeData V1 paper: Read as text on 8 Nov 2026.
- BridgeData V2 project page: Read as text on 8 Nov 2026.
- BridgeData V2 paper: Read as text on 8 Nov 2026.
- CALVIN paper: Read as text on 8 Nov 2026.
- CALVIN repository: Read as text on 8 Nov 2026.
- D3IL paper: Read as text on 8 Nov 2026.
- D3IL repository: Read as text on 8 Nov 2026.
- DROID project page: Read as text on 8 Nov 2026.
- DROID paper: Read as text on 8 Nov 2026.
- DROID project page: Read as text on 8 Nov 2026.
- FMB paper: Read as text on 8 Nov 2026.
- FMB project page: Read as text on 8 Nov 2026.
- FurnitureBench project page: Read as text on 8 Nov 2026.
- FurnitureBench paper: Read as text on 8 Nov 2026.
- FurnitureBench repository: Read as text on 8 Nov 2026.
- Homes of New York paper: Read as text on 8 Nov 2026.
- Homes of New York repository: Read as text on 8 Nov 2026.
- Humanoid Everyday paper: Read as text on 8 Nov 2026.
- LIBERO paper: Read as text on 8 Nov 2026.
- LIBERO repository: Read as text on 8 Nov 2026.
- ManiSkill2 paper: Read as text on 8 Nov 2026.
- ManiSkill2 repository: Read as text on 8 Nov 2026.
- MIME paper: Read as text on 8 Nov 2026.
- Open X-Embodiment project page: Read as text on 8 Nov 2026.
- Open X-Embodiment paper: Read as text on 8 Nov 2026.
- Open X-Embodiment licence file: Read as text on 8 Nov 2026.
- Real Robot Challenge datasets paper: Read as text on 8 Nov 2026.
- Real Robot Challenge datasets repository: Read as text on 8 Nov 2026.
- RH20T project page: Read as text on 8 Nov 2026.
- RLBench paper: Read as text on 8 Nov 2026.
- RLBench licence file: Read as text on 8 Nov 2026.
- RoboCasa paper: Read as text on 8 Nov 2026.
- RoboCasa project page: Read as text on 8 Nov 2026.
- RoboCasa repository: Read as text on 8 Nov 2026.
- robomimic project page: Read as text on 8 Nov 2026.
- robomimic paper: Read as text on 8 Nov 2026.
- robomimic repository: Read as text on 8 Nov 2026.
- RoboNet paper: Read as text on 8 Nov 2026.
- RoboNet project page: Read as text on 8 Nov 2026.
- RoboNet licence file: Read as text on 8 Nov 2026.
- RoboSet project page: Read as text on 8 Nov 2026.
- RoboSet paper: Read as text on 8 Nov 2026.
- RoboTurk pilot (simulated) paper: Read as text on 8 Nov 2026.
- RoboTurk pilot (simulated) project page: Read as text on 8 Nov 2026.
- RoboTurk real-world project page: Read as text on 8 Nov 2026.
- RoboTurk real-world project page: Read as text on 8 Nov 2026.
- Universal Manipulation Interface paper: Read as text on 8 Nov 2026.
- Universal Manipulation Interface project page: Read as text on 8 Nov 2026.
- Universal Manipulation Interface repository: Read as text on 8 Nov 2026.
Continue this reading path
- Best Robot Training Data category guide
- How to choose: a robot training data provider
- Resource directory
Cite this entry
Best Robot Training Data. "Open robot dataset spec ledger." Content reviewed 8 November 2026. https://bestrobottrainingdata.com/dataset-spec-ledger/
General information only. This is not legal advice. Privacy and data rights vary with where and how data is collected, so talk to qualified counsel before collecting or buying data that shows people.