robotindex Hugging Face Robotics Datasets
Summary · 2 October 2026 · 100+ monthly downloads
1,467
datasets tracked
5,351,203
monthly downloads
937
created in the last 90 days
503
robots identified
Most used
By downloads in the last 30 days.
| # | Downloads | Likes | Age | Dataset | License | Robot |
|---|---|---|---|---|---|---|
| 1 | 1,222,736 | 275 | 563d | PhysicalAI-Robotics-GR00T-X-Robots-SimnvidiaPost-training datasets for GR00T N1, spanning many robot embodiments in simulation. | cc-by-4.0 | Unitree G1 |
| 2 | 291,875 | 50 | 81d | Gen-HumanEgogenrobot20251,800+ hours of egocentric human demonstrations with synchronised structured supervision. | cc-by-sa-4.0 | human video |
| 3 | 258,951 | 48 | 66d | ACE-Data-0ACERoboticsAmbient human-centric capture as an embodied data engine, from NTU Singapore's S-Lab. | other | human video |
| 4 | 252,657 | 12 | 57d | challenge_datachallenge-2026PrimeBot household bimanual manipulation challenge — real teleoperation, train and validation splits. | cc-by-sa-4.0 | UMI (handheld) |
| 5 | 240,794 | 127 | 109d | ABC-130kXDOFDescribed as the largest open-source robot teleoperation dataset: bimanual trajectories on two-arm YAM. | apache-2.0 | YAM |
| 6 | 211,548 | 3 | 64d | stereo-550fpvlabsStereo vision data collected on FPV Labs' open-source hardware, with a build guide and 3D viewer. | other | human video |
| 7 | 188,673 | 55 | 238d | PhysicalAI-Robotics-Open-H-RobotsnvidiaCommunity-driven effort to build an open shared foundation for humanoid training data. | cc-by-4.0 | — |
| 8 | 173,110 | 143 | 302d | Egocentric-100KbuilddotaiThe largest dataset of manual labour captured from a first-person view. | apache-2.0 | human video |
| 9 | 155,205 | 52 | 115d | HIW-500BitRobotHumanoids In-the-Wild — large-scale whole-body humanoid data recorded outside the lab. | cc-by-4.0 | Unitree G1 |
| 10 | 114,549 | 56 | 75d | HiFi-UMI-2Ksimple-world-lab2,000 hours of robot-free manipulation via handheld UMI grippers, six synchronised camera views, 480+ scenes. | cc-by-4.0 | UMI (handheld) |
| 11 | 101,430 | 6 | 396d | TartanGroundtheairlabcmuLarge-scale ground robot perception and navigation data from CMU's AirLab. | cc-by-4.0 | — |
| 12 | 84,056 | 105 | 56d | EgoDemoLightwheelAIA 50-hour sample of EgoSuite-Open100K covering every annotated subset plus raw video. | other | — |
| 13 | 82,729 | 249 | 205d | xperience-10mropedia-aiLarge recent release behind an access request; terms not confirmed. | other | human video |
| 14 | 79,286 | 286 | 339d | OmniActionOpenMOSS-TeamRoboOmni — proactive robot manipulation in an omni-modal context, accepted to ICLR 2026. | cc-by-nc-4.0 | — |
| 15 | 78,776 | 11 | 94d | 2026-challenge-demosbehavior-1kDemonstration data for the BEHAVIOR-1K 2026 challenge. | mit | — |
| 16 | 78,682 | 24 | 113d | Hy-Embodied-0.5-VLA-DatatencentTencent's vision-language-action training data, part of a real-world robot learning stack. | cc-by-4.0 | — |
| 17 | 51,875 | 8 | 211d | molmobot-dataallenaiTraining episodes for eight tasks across two robot platforms, from AI2's MolmoBot. | odc-by | — |
| 18 | 50,428 | 54 | 638d | RoboMINDx-humanoid-robomindMulti-robot benchmark of normative data for robot manipulation. | apache-2.0 | — |
| 19 | 47,678 | 10 | 119d | QueST-PartNetMobility-SAPIENAnandMayankPartNet-Mobility assets simulated in SAPIEN, accompanying the QueST paper. | cc-by-4.0 | — |
| 20 | 35,382 | 110 | 56d | EgoProLightwheelAIThe 10,000-hour head-and-wrist line of EgoSuite-Open100K. | other | — |
New
Created in the last 90 days, ranked by downloads.
| # | Downloads | Per day | Age | Dataset | License | Robot |
|---|---|---|---|---|---|---|
| 1 | 291,875 | 3,603 | 81d | Gen-HumanEgogenrobot20251,800+ hours of egocentric human demonstrations with synchronised structured supervision. | cc-by-sa-4.0 | human video |
| 2 | 258,951 | 3,924 | 66d | ACE-Data-0ACERoboticsAmbient human-centric capture as an embodied data engine, from NTU Singapore's S-Lab. | other | human video |
| 3 | 252,657 | 4,433 | 57d | challenge_datachallenge-2026PrimeBot household bimanual manipulation challenge — real teleoperation, train and validation splits. | cc-by-sa-4.0 | UMI (handheld) |
| 4 | 211,548 | 3,305 | 64d | stereo-550fpvlabsStereo vision data collected on FPV Labs' open-source hardware, with a build guide and 3D viewer. | other | human video |
| 5 | 114,549 | 1,527 | 75d | HiFi-UMI-2Ksimple-world-lab2,000 hours of robot-free manipulation via handheld UMI grippers, six synchronised camera views, 480+ scenes. | cc-by-4.0 | UMI (handheld) |
| 6 | 84,056 | 1,501 | 56d | EgoDemoLightwheelAIA 50-hour sample of EgoSuite-Open100K covering every annotated subset plus raw video. | other | — |
| 7 | 35,382 | 632 | 56d | EgoProLightwheelAIThe 10,000-hour head-and-wrist line of EgoSuite-Open100K. | other | — |
| 8 | 34,874 | 811 | 43d | Daimon-Infinityml-resourcesA file-preserving mirror of Daimon Robotics' dataset, originally published on ModelScope. | cc-by-nc-sa-4.0 | — |
| 9 | 31,321 | 979 | 32d | UniPhys-40KbreezexianUnified physical grounding for heterogeneous 3D assets, with articulation and material properties. | other | — |
| 10 | 27,799 | 1,264 | 22d | World-SimReady-HomeYoottaCAD-based SimReady home assets with configured collisions, built for simulation. | cc-by-nc-sa-4.0 | — |
| 11 | 23,422 | 520 | 45d | robotwin_3dflex-piBimanual manipulation from the RoboTwin 2.0 simulator with RGB and depth, in LeRobot v2.1 format. | — | ALOHA |
| 12 | 23,196 | 273 | 85d | UrbanVerse-Training-ScenesUCLA-VAIL | cc-by-4.0 | — |
| 13 | 19,395 | 313 | 62d | hapticwam-teleop-rawarmteamRaw teleoperated episodes behind HapticWAM, a world-action model trained without real tactile sensing. | cc-by-4.0 | — |
| 14 | 18,463 | 293 | 63d | airoa-moma-5kairoa-org | other | — |
| 15 | 18,042 | 228 | 79d | OpenNeoDataNeoteAIEmbodied | cc-by-nc-sa-4.0 | — |
| 16 | 16,651 | 979 | 17d | EgoSteer-RealWorldEgoSteerReal-robot bimanual teleoperation collected for the EgoSteer project. | apache-2.0 | human video |
| 17 | 15,424 | 617 | 25d | Public-YAM-runsandlyuBimanual YAM episodes recorded on the BluPe operator station, one episode added per run. | — | YAM |
| 18 | 12,068 | 157 | 77d | DataDump_07-17-2026RoboArena | mit | — |
| 19 | 11,246 | 803 | 14d | hoi-retargetleggedroboticsHumanoid human-object interaction motion, retargeted for learning from demonstration. | cc-by-nc-sa-4.0 | Unitree G1Unitree H2motion capture |
| 20 | 10,571 | 320 | 33d | RoboPulsePlusPluslyy0715Interval-level benchmark from the PRM-as-a-Judge toolkit. | other | — |
Trending
Downloads per day weighted toward recent releases. A heuristic on a single snapshot — real week-on-week change begins with the next harvest.
| # | Trend | Per day | Age | Dataset | License | Robot |
|---|---|---|---|---|---|---|
| 1 | 2,757 | 4,433 | 57d | challenge_datachallenge-2026PrimeBot household bimanual manipulation challenge — real teleoperation, train and validation splits. | cc-by-sa-4.0 | UMI (handheld) |
| 2 | 2,264 | 3,924 | 66d | ACE-Data-0ACERoboticsAmbient human-centric capture as an embodied data engine, from NTU Singapore's S-Lab. | other | human video |
| 3 | 1,939 | 3,305 | 64d | stereo-550fpvlabsStereo vision data collected on FPV Labs' open-source hardware, with a build guide and 3D viewer. | other | human video |
| 4 | 1,834 | 3,603 | 81d | Gen-HumanEgogenrobot20251,800+ hours of egocentric human demonstrations with synchronised structured supervision. | cc-by-sa-4.0 | human video |
| 5 | 1,052 | 1,264 | 22d | World-SimReady-HomeYoottaCAD-based SimReady home assets with configured collisions, built for simulation. | cc-by-nc-sa-4.0 | — |
| 6 | 941 | 1,501 | 56d | EgoDemoLightwheelAIA 50-hour sample of EgoSuite-Open100K covering every annotated subset plus raw video. | other | — |
| 7 | 891 | 2,209 | 109d | ABC-130kXDOFDescribed as the largest open-source robot teleoperation dataset: bimanual trajectories on two-arm YAM. | apache-2.0 | YAM |
| 8 | 850 | 979 | 17d | EgoSteer-RealWorldEgoSteerReal-robot bimanual teleoperation collected for the EgoSteer project. | apache-2.0 | human video |
| 9 | 817 | 1,527 | 75d | HiFi-UMI-2Ksimple-world-lab2,000 hours of robot-free manipulation via handheld UMI grippers, six synchronised camera views, 480+ scenes. | cc-by-4.0 | UMI (handheld) |
| 10 | 750 | 979 | 32d | UniPhys-40KbreezexianUnified physical grounding for heterogeneous 3D assets, with articulation and material properties. | other | — |
| 11 | 715 | 803 | 14d | hoi-retargetleggedroboticsHumanoid human-object interaction motion, retargeted for learning from demonstration. | cc-by-nc-sa-4.0 | Unitree G1Unitree H2motion capture |
| 12 | 567 | 811 | 43d | Daimon-Infinityml-resourcesA file-preserving mirror of Daimon Robotics' dataset, originally published on ModelScope. | cc-by-nc-sa-4.0 | — |
| 13 | 518 | 1,350 | 115d | HIW-500BitRobotHumanoids In-the-Wild — large-scale whole-body humanoid data recorded outside the lab. | cc-by-4.0 | Unitree G1 |
| 14 | 501 | 617 | 25d | Public-YAM-runsandlyuBimanual YAM episodes recorded on the BluPe operator station, one episode added per run. | — | YAM |
| 15 | 396 | 632 | 56d | EgoProLightwheelAIThe 10,000-hour head-and-wrist line of EgoSuite-Open100K. | other | — |
| 16 | 383 | 838 | 94d | 2026-challenge-demosbehavior-1kDemonstration data for the BEHAVIOR-1K 2026 challenge. | mit | — |
| 17 | 357 | 520 | 45d | robotwin_3dflex-piBimanual manipulation from the RoboTwin 2.0 simulator with RGB and depth, in LeRobot v2.1 format. | — | ALOHA |
| 18 | 327 | 364 | 13d | TraceSpatial-TraceVCG-EAIThe RGB and question-answer tracing subset of TraceSpatial. | apache-2.0 | AgiBot |
| 19 | 271 | 696 | 113d | Hy-Embodied-0.5-VLA-DatatencentTencent's vision-language-action training data, part of a real-world robot learning stack. | cc-by-4.0 | — |
| 20 | 243 | 320 | 33d | RoboPulsePlusPluslyy0715Interval-level benchmark from the PRM-as-a-Judge toolkit. | other | — |
Dataset categories
| Category | Datasets | |
|---|---|---|
| Manipulation → | 635 | |
| Simulation → | 162 | |
| Other → | 162 | |
| Teleoperation → | 155 | |
| Benchmark & Eval → | 76 | |
| Egocentric & Human → | 72 | |
| Navigation & Driving → | 60 | |
| Locomotion → | 57 | |
| Perception → | 56 | |
| Language & VLA → | 30 | |
| Motion Capture → | 2 |
Robots
Robots you can buy today. The 200 most-downloaded datasets (59 with identifiable hardware) against all 1,467 (503). Discontinued platforms are excluded and listed below.
| Platform | Top 200 | All 1,467 | ||
|---|---|---|---|---|
| SO-100/101 ↗ | 2 | 64 | ||
| Unitree G1 ↗ | 8 | 60 | ||
| UR5 ↗ | 6 | 58 | ||
| AgileX PiPER ↗ | 7 | 44 | ||
| ALOHA ↗ | 8 | 38 | ||
| xArm ↗ | 2 | 17 | ||
| AgiBot ↗ | 7 | 16 | ||
| Kuka ↗ | 2 | 16 | ||
| YAM ↗ | 3 | 15 | ||
| Galaxea ↗ | 2 | 14 | ||
| Fourier GR-1 ↗ | 2 | 12 | ||
| WidowX ↗ | 4 | 11 | ||
| Pollen Reachy ↗ | 2 | 11 | ||
| UMI (handheld) ↗ | 2 | 10 | ||
| Unitree H1 | 1 | 9 | ||
| myCobot ↗ | 3 | 5 |
Discontinued
Excluded above. The data remains useful; the robot is not purchasable as recorded.
| Platform | Top 200 | All | Status |
|---|---|---|---|
| Franka Panda ↗ | 20 | 155 | Panda withdrawn; FR3 is the current model. Most cards say only "Franka". |
| Kinova Jaco ↗ | — | 9 | Listed discontinued by the editor; Kinova has not confirmed and the Gen2 is still resold |
| Sawyer ↗ | — | 5 | Rethink Robotics bankrupt 2018; relaunch closed 2024. |