datasets
Training and evaluation data, with the modality, task and licence stated up front. Listed live from the Hugging Face Hub.
LA_dataset_human_made_franka_eef_Lerobotv21_260511train_ctf_eefrobotwin_v3_eef
RoboTwin v3 EEF (LeRobot v3)
This is a LeRobot v3 format conversion of the RoboTwin bimanual benchmark (source: robbyant/robotwin-clean-and-aug-lerobot), reorganized into a canonical bimanual 16D end-effector (EE) pose representation. 50 Hz, 50 tasks, two subsets (clean/ 50 demos/task and aug/ 500 demos/task).
Original Dataset
RoboTwin 2.0: A Scalable Data Generator and Benchmark with Strong Domain Randomization for Robust Bimanual Robotic Manipulation
Chen, T.… See the full description on the dataset page: https://huggingface.co/datasets/GT-111/robotwin_v3_eef.intern_a1_v3_eef
InternData-A1 v3 EEF (LeRobot v3)
This is a LeRobot v3 format conversion of InternData-A1, reorganized into a canonical bimanual 16D end-effector (EE) pose representation for large-scale VLA pre-training. 30 Hz, four embodiments, ~487.7K episodes and ~377.7M frames.
Original Dataset
InternData-A1: Pioneering High-Fidelity Synthetic Data for Pre-training Generalist Policy
InternRobotics. InternData-A1: Pioneering High-Fidelity Synthetic Data for Pre-training… See the full description on the dataset page: https://huggingface.co/datasets/GT-111/intern_a1_v3_eef.robomme_v3_eef
RoboMME v3 EEF (LeRobot v3)
This is a LeRobot v3 format conversion of RoboMME, the first large-scale robotic benchmark dedicated to memory-augmented manipulation, reorganized into a canonical 8D end-effector (EE) pose representation. 20 Hz, single-arm Franka Panda, with cognitively motivated task suites covering pick-and-place, routing, pattern memory, and video-conditioned tasks.
Original Dataset
RoboMME: A Robotic Benchmark for Memory-Augmented Manipulation… See the full description on the dataset page: https://huggingface.co/datasets/GT-111/robomme_v3_eef.agibotworld_alpha_v3_eef
AgiBotWorld-Alpha (LeRobot v3 EE)
This is a LeRobot v3 format conversion of the AgiBotWorld-Alpha dataset, reorganized into a canonical bimanual 16D end-effector (EE) pose representation for multi-dataset robot learning. 30 Hz, 33,609 episodes, ~56.9M frames.
Original Dataset
AgiBot World Colosseo: A Large-scale Manipulation Platform for Scalable and Intelligent Embodied Systems
Bu, Q., Cai, J., Chen, L., et al. (AgiBot-World-Contributors). AgiBot World Colosseo:… See the full description on the dataset page: https://huggingface.co/datasets/GT-111/agibotworld_alpha_v3_eef.play_robot_fruits_train_eefThis dataset was created using LeRobot.
Dataset Structure
meta/info.json:
{
"codebase_version": "v2.1",
"trossen_subversion": "v1.0",
"robot_type": "trossen_ai_stationary",
"total_episodes": 250,
"total_frames": 263356,
"total_tasks": 10,
"total_videos": 500,
"total_chunks": 1,
"chunks_size": 1000,
"fps": 30,
"splits": {
"train": "0:250"
},
"data_path":… See the full description on the dataset page: https://huggingface.co/datasets/aytsaiusc/play_robot_fruits_train_eef.sanitizer-pick-place-eef
Tuesday-Lab/sanitizer-pick-place-eef
LeRobot-format robot dataset uploaded from a local dataset root.
Metadata
codebase_version: v2.1
fps: 50
robot_type: tuesday_arm
Load Example
from lerobot.datasets.lerobot_dataset import LeRobotDataset
dataset = LeRobotDataset(
repo_id="Tuesday-Lab/sanitizer-pick-place-eef",
root=None, # Cache path; keep None to use default HF cache
video_backend="pyav",
)
print(len(dataset))
play_human_fruits_v2_train_eef
play_human_fruits_v2_train_eef
Human EEF annotations for aytsaiusc/play_human_fruits_train.
2026-08-26: rebuilt after fixing a hand-detection defect. If you pulled this dataset before
that date, re-pull — the earlier build annotated only 87.6% of frames (worst episode 50.9%) and
its hand boxes were systematically cut off.
What was wrong with the earlier build
Hand detection ran Grounding-DINO on a hard-coded workspace window whose right edge sat at
x=440. The hand… See the full description on the dataset page: https://huggingface.co/datasets/yusenluo9z/play_human_fruits_v2_train_eef.hil_pi05_eef_100kThis dataset was created using LeRobot.
Dataset Structure
meta/info.json:
{
"codebase_version": "v3.0",
"fps": 50,
"features": {
"observation.state": {
"dtype": "float32",
"shape": [
13
],
"names": [
"eef.x.pos",
"eef.y.pos",
"eef.z.pos",
"eef.xaxis.x.pos",
"eef.xaxis.y.pos",
"eef.xaxis.z.pos"… See the full description on the dataset page: https://huggingface.co/datasets/maskjp/hil_pi05_eef_100k.anubis_stack_tuna_eefpose_50_v2_1This dataset was created using LeRobot.
Dataset Structure
meta/info.json:
{
"codebase_version": "v2.1",
"robot_type": null,
"total_episodes": 50,
"total_frames": 11830,
"total_tasks": 1,
"total_videos": 0,
"total_chunks": 1,
"chunks_size": 1000,
"fps": 50,
"splits": {
"train": "0:50"
},
"data_path": "data/chunk-{episode_chunk:03d}/episode_{episode_index:06d}.parquet",
"video_path": null,
"features": {… See the full description on the dataset page: https://huggingface.co/datasets/chomeed/anubis_stack_tuna_eefpose_50_v2_1.Robocasa_Camera_Space_Eef
RoboCasa Camera-Space EEF (ACE-Ego-0)
LeRobot v2.0 demonstrations for the 24 GR1 RoboCasa TableTop tasks used by
ACE-Ego-0 SFT. Actions and states
are stored in the observation camera frame (end-effector position + rot6d, plus
gripper and waist).
This dataset does not include unused conversion caches (experiments/) or
legacy per-dataset index files. Training reads only the files listed below and
the shipped merged q99 statistics. The training code never recomputes norms.… See the full description on the dataset page: https://huggingface.co/datasets/ACERobotics/Robocasa_Camera_Space_Eef.cola_firm_eef_raw_5hzsanitizer-pick-place-openpi-eef-v21
Tuesday-Lab/sanitizer-pick-place-openpi-eef-v21
LeRobot-format robot dataset uploaded from a local dataset root.
Metadata
codebase_version: v2.1
fps: 50
robot_type: tuesday_arm
Load Example
from lerobot.datasets.lerobot_dataset import LeRobotDataset
dataset = LeRobotDataset(
repo_id="Tuesday-Lab/sanitizer-pick-place-openpi-eef-v21",
root=None, # Cache path; keep None to use default HF cache
video_backend="pyav",
)
print(len(dataset))
catheter_force_depth_eef
catheter_force_depth_eef
This dataset was created using LeRobot and is stored in the episode-split LeRobot v2.1 layout used by OpenPI.
Dataset Structure
The dataset uses parquet files for per-frame robot observations/actions and episode-split videos for camera streams.
meta/
info.json
tasks.jsonl
episodes.jsonl
stats.json
episodes_stats.jsonl
data/
chunk-000/
episode_000000.parquet
videos/
chunk-000/
<video_key>/
episode_000000.mp4… See the full description on the dataset page: https://huggingface.co/datasets/giakhuyendihoc/catheter_force_depth_eef.real_world_maniskill_lift_peg_eefplay_human_fruits_v2_fix_train_eef
play_human_fruits_v2_fix_train_eef
Human EEF annotations for aytsaiusc/play_human_fruits_train, rebuilt after fixing a hand
detection defect. This supersedes yusenluo9z/play_human_fruits_v2_train_eef — use this one.
What was wrong with the previous version
Hand detection ran Grounding-DINO on a hard-coded workspace window whose right edge sat at
x=440. The hand enters this rig from the right, so the window cut it in half: the detector only
ever saw the fingertips… See the full description on the dataset page: https://huggingface.co/datasets/yusenluo9z/play_human_fruits_v2_fix_train_eef.libero_v3_eef
LIBERO Combined 4x10 — canonical EEF v2
This repository contains 1,693 oracle episodes for the 40 LIBERO evaluation
tasks at their native 10 Hz. The persisted action contract is
eef_absolute_next_observation_wxyz_v2.
State and action
observation.state and action are both
[x, y, z, qw, qx, qy, qz, gripper]: metres, scalar-first WXYZ quaternion,
and symmetric finger openness in metres. On every row where
transition_action_valid=true, action[t] equals the next… See the full description on the dataset page: https://huggingface.co/datasets/GT-111/libero_v3_eef.echelon-original-eef-main_right_wristThis dataset was created using LeRobot.
Dataset Structure
meta/info.json:
{
"codebase_version": "v2.0",
"robot_type": "ur3e",
"total_episodes": 193,
"total_frames": 5829,
"total_tasks": 13,
"total_videos": 0,
"total_chunks": 1,
"chunks_size": 1000,
"fps": 5,
"splits": {
"train": "0:193"
},
"data_path": "data/chunk-{episode_chunk:03d}/episode_{episode_index:06d}.parquet",
"video_path":… See the full description on the dataset page: https://huggingface.co/datasets/microsoft/echelon-original-eef-main_right_wrist.base4-mobile-door-eef-merged-v30This dataset was created using LeRobot.
Dataset Structure
meta/info.json:
{
"codebase_version": "v3.0",
"fps": 50,
"features": {
"observation.state": {
"dtype": "float32",
"shape": [
10
],
"names": {
"motors": [
"joint1",
"joint2",
"joint3",
"joint4",
"joint5"… See the full description on the dataset page: https://huggingface.co/datasets/L5vel/base4-mobile-door-eef-merged-v30.ww_dataset_zero_delta_eefThis dataset was created using LeRobot.
Dataset Structure
meta/info.json:
{
"codebase_version": "v2.1",
"robot_type": "ur5",
"total_episodes": 200,
"total_frames": 29373,
"total_tasks": 2,
"total_videos": 400,
"total_chunks": 1,
"chunks_size": 1000,
"fps": 10,
"splits": {
"train": "0:200"
},
"data_path": "data/chunk-{episode_chunk:03d}/episode_{episode_index:06d}.parquet",
"video_path":… See the full description on the dataset page: https://huggingface.co/datasets/pmoller/ww_dataset_zero_delta_eef.train_eefclean_up_table_eef32This dataset was created using LeRobot.
Dataset Structure
meta/info.json:
{
"codebase_version": "v2.1",
"robot_type": "yam_bimanual",
"total_episodes": 308,
"total_frames": 221680,
"total_tasks": 6,
"total_videos": 924,
"total_chunks": 1,
"chunks_size": 1000,
"fps": 30,
"splits": {
"train": "0:308"
},
"data_path": "data/chunk-{episode_chunk:03d}/episode_{episode_index:06d}.parquet",
"video_path":… See the full description on the dataset page: https://huggingface.co/datasets/YOLO2431/clean_up_table_eef32.anubis_pick_carrot_out_of_pot_eefpose_v2_1This dataset was created using LeRobot.
Dataset Structure
meta/info.json:
{
"codebase_version": "v2.1",
"robot_type": null,
"total_episodes": 70,
"total_frames": 28868,
"total_tasks": 1,
"total_videos": 210,
"total_chunks": 1,
"chunks_size": 1000,
"fps": 50,
"splits": {
"train": "0:70"
},
"data_path": "data/chunk-{episode_chunk:03d}/episode_{episode_index:06d}.parquet",
"video_path":… See the full description on the dataset page: https://huggingface.co/datasets/chomeed/anubis_pick_carrot_out_of_pot_eefpose_v2_1.eef_motionThis dataset was created using LeRobot.
Dataset Structure
meta/info.json:
{
"codebase_version": "v3.0",
"fps": 20,
"features": {
"observation.images.robot0_eye_in_hand": {
"dtype": "video",
"shape": [
256,
256,
3
],
"names": [
"height",
"width",
"channel"
],
"video_info": {… See the full description on the dataset page: https://huggingface.co/datasets/euijinrnd/eef_motion.ww_dataset_abs_joint_delta_eefThis dataset was created using LeRobot.
Dataset Structure
meta/info.json:
{
"codebase_version": "v2.1",
"robot_type": "ur5",
"total_episodes": 200,
"total_frames": 29373,
"total_tasks": 2,
"total_videos": 400,
"total_chunks": 1,
"chunks_size": 1000,
"fps": 10,
"splits": {
"train": "0:200"
},
"data_path": "data/chunk-{episode_chunk:03d}/episode_{episode_index:06d}.parquet",
"video_path":… See the full description on the dataset page: https://huggingface.co/datasets/pmoller/ww_dataset_abs_joint_delta_eef.rlbench_v3_eef
RLBench-18 v3 EEF — Train Split (LeRobot v3.0)
RLBench 18-task PerAct dataset converted to LeRobot v3.0 with lossless H.264 video (CRF=0, yuv444p).
Format
Per-task LeRobot v3.0 datasets — each task is a subdirectory with its own meta/, data/, videos/
8D EEF: xyz + quaternion (wxyz, scalar-first) + gripper
State: absolute pose (wxyz passthrough from PerAct)
Action: delta position + absolute quaternion (wxyz) + gripper
2 cameras: front view… See the full description on the dataset page: https://huggingface.co/datasets/GT-111/rlbench_v3_eef.base4-clean-table-eef-merged-v30This dataset was created using LeRobot.
Dataset Structure
meta/info.json:
{
"codebase_version": "v3.0",
"fps": 50,
"features": {
"observation.state": {
"dtype": "float32",
"shape": [
10
],
"names": {
"motors": [
"joint1",
"joint2",
"joint3",
"joint4",
"joint5"… See the full description on the dataset page: https://huggingface.co/datasets/L5vel/base4-clean-table-eef-merged-v30.robotwin_eef_clean_500alex_open_door_eef_1