Robot learning dataset
DROID
DROID is a real-world manipulation dataset collected with Franka Panda arms across 564 scenes. The project reports 76k demonstration trajectories and 350 hours, and distributes RLDS training data and higher-resolution raw recordings. A dataset-wide license was not found in the opened official documentation.
Records last verified Oct 13, 2026 · Send a correction
Licence & use
Licence not established
Commercial use: not established
Terms apply to the checked release. Read publisher conditions for mixed collections, converted formats and gated access.
- Publisher
- DROID project
- Released
- Not dated
- Published size
- 350 h · 76,000 trajectories · 1.7 TB
- Modalities
- rgb, proprioception, language
- Formats
- RLDS, TFDS, MP4
- Real / simulated
- Real
- Includes humanoid recordings
- No
Size & release notes
Project publishes 76k trajectories and 350 hours. Official docs list 1.7 TB for full RLDS data, 8.7 TB for stereo-HD raw recordings, and 5.6 TB when excluding stereo videos and SVO files. These are alternative representations, not additive sizes. Raw data includes stereo/depth information.
Recorded embodiments
- Franka Panda
Models disclosing this training dataset
More robot learning datasets
Evidence
Sources
- [s1] DROID project page ↗
DROID project · official · accessed Oct 13, 2026
- [s2] DROID robot platform repository ↗
DROID project · official · accessed Oct 13, 2026
- [s3] DROID: A Large-Scale In-The-Wild Robot Manipulation Dataset ↗
arXiv · paper · published Mar 19, 2024 · accessed Oct 13, 2026
- [s4] The DROID Dataset: download formats and RLDS schema ↗
DROID project · official · accessed Oct 13, 2026