Datasets

Browse a curated catalog of datasets for humanoid robots and embodied AI, spanning real-world demonstrations, teleoperation, motion capture, egocentric vision, simulation, manipulation, locomotion, and cross-embodiment robot learning.

83 Datasets · Page 1 of 7

AIRoA MoMa 5k

AIRoA MoMa 5k contains 1,184,259 successful primitive-action episodes spanning 5,025.1 hours and 180,905,084 frames of teleoperated mobile manipulation with 44 Toyota Human Support Robots across five sites. The public, train-only LeRobot v3.0 release covers 68 main short-horizon task templates and combines head- and hand-camera RGB video with robot state and actions, end-effector pose, wrist force-torque history, low-level servo telemetry, and hierarchical task metadata.

View dataset →

A mosaic of RealOmni household manipulation demonstrations

10Kh RealOmni-Open Dataset

The 10Kh RealOmni-Open Dataset contains more than 13,000 hours and 5 million clips of bimanual human demonstrations captured across over 10,000 real household scenarios. Collected from more than 3,000 contributors with GenDAS grippers, it covers 30 manipulation skills in 10 scenario groups, including clothing, clutter organization, kitchen cleaning, and shoe handling. The 95 TB MCAP release pairs 1600 × 1296 fisheye video at 30 fps with reconstructed end-effector trajectories, gripper state, 6-axis IMU readings, and tactile-array signals.

View dataset →

A bimanual dexterous robot folding a paper airplane

Robotic Origami Challenge

The Robotic Origami Challenge dataset contains 682 real-world teleoperated demonstrations of a bimanual dexterous robot folding a traditional six-fold paper airplane. Its 51 collection seasons comprise more than 4.7 million frames at 30 fps, with 65-dimensional robot states and actions, joint torques, 10-fingertip force-torque signals, and six synchronized head, wrist, and tactile video streams. The release uses LeRobot v3.0, with many seasons also available in v2.1, for imitation learning, visual-tactile representation learning, and long-horizon deformable-object manipulation.

View dataset →

A mosaic of synchronized HiFi-UMI manipulation demonstrations

HiFi-UMI-2K

HiFi-UMI-2K is a 2,000-hour release of robot-free, bimanual manipulation demonstrations collected with the portable HiFi-UMI system. Each episode includes six microsecond-synchronized camera views, calibrated bimanual trajectories, gripper states, language annotations, subtask boundaries, and per-sample quality metadata. The data is automatically reconstructed and validated through simulation replay, then exported in a LeRobot v3-style Parquet and MP4 format for manipulation-policy pre-training and post-training.

View dataset →

EPIC-KITCHENS-100

EPIC-KITCHENS-100 contains 100 hours and 20 million frames of unscripted first-person activity recorded in 45 kitchens across 700 videos. Its 89,900 action segments and 20,500 multilingual narrations support action recognition, detection, anticipation, domain adaptation, and video-text retrieval.

View dataset →

Ego4D

Ego4D contains more than 3,670 hours of first-person daily-life video from 923 participants across 74 locations in nine countries. Portions include audio, narrations, gaze, stereo video, synchronized wearable views, and 3D scans, with benchmarks for episodic memory, hand-object interaction, social understanding, diarization, and forecasting.

View dataset →

Ego-Exo4D

Ego-Exo4D captures 1,286.3 hours of synchronized first- and third-person video across 5,035 takes, 740 participants, and eight skilled-activity domains. Aria glasses and multiple external cameras provide calibrated video, audio, eye gaze, inertial data, trajectories, point clouds, and three forms of language annotation for skill and cross-view understanding.

View dataset →

DexYCB

DexYCB contains 582,000 RGB-D frames from 1,000 sequences of 10 people grasping 20 YCB objects, recorded synchronously from eight viewpoints. It provides calibrated imagery and 3D hand and object poses for hand-pose estimation, 6D object-pose estimation, and safe human-to-robot handover research.

View dataset →

robomimic

robomimic combines proficient-human, multi-human, and machine-generated demonstrations for five simulated and three real robot manipulation tasks. The HDF5 datasets include low-dimensional or multi-camera image observations for pick-and-place, insertion, tool hanging, and bimanual transport, supporting reproducible imitation and offline reinforcement learning.

View dataset →

CALVIN

CALVIN contains 24 hours of teleoperated play data, with six hours in each of four simulated tabletop environments and roughly 20,000 language directives. The benchmark evaluates language-conditioned policies on long sequences of object manipulation using static and gripper cameras, proprioception, depth, or tactile observations.

View dataset →

ManiSkill2

ManiSkill2 provides more than four million demonstration frames for 20 simulated manipulation skill families and over 2,000 object models. Its SAPIEN environments span stationary and mobile platforms, single- and dual-arm control, rigid and soft bodies, RGB-D and point-cloud observations, and tasks from picking and insertion to pouring and excavation.

View dataset →

ALOHA Static

ALOHA Static comprises 825 real demonstrations of fine-grained bimanual tabletop manipulation across 12 tasks, including sealing a bag, wrapping candy, opening a cup, preparing tape, using a coffee machine, fastening cable, slotting a battery, and handing over tools. The fixed ALOHA workcell records 14-dimensional joint state and action data at 50 Hz alongside four 480×640 RGB views from two stationary and two wrist-mounted cameras. The task-specific releases are distributed in LeRobot format for imitation-learning research and co-training.

View dataset →