datasets
Training and evaluation data, with the modality, task and licence stated up front. Listed live from the Hugging Face Hub.
ABC-130k
ABC-130k
ABC-130k is the largest open-source robot teleoperation dataset. It contains
bimanual manipulation trajectories collected on two-arm YAM stations. Episodes
are distributed as MCAP files, with subtask annotations kept as separate
artifacts so they can be revised or extended independently of the underlying
episode data. For details on the accompanying paper, see abc.bot.
Please see the GitHub repo here for code to
train and deploy with this dataset.
Dataset… See the full description on the dataset page: https://huggingface.co/datasets/XDOF/ABC-130k.abc_130k_v3_trainThis dataset was created using LeRobot.
Dataset Structure
meta/info.json:
{
"codebase_version": "v3.0",
"fps": 30,
"features": {
"observation.state": {
"dtype": "float32",
"shape": [
14
],
"names": [
"left_arm_joint_1",
"left_arm_joint_2",
"left_arm_joint_3",
"left_arm_joint_4",
"left_arm_joint_5"… See the full description on the dataset page: https://huggingface.co/datasets/lerobot/abc_130k_v3_train.abc130k_v3_train
ABC-130k — 256² full-frame re-conversion (train split)
Full-resolution re-conversion of the ABC-130k dual-YAM bimanual manipulation corpus
(XDOF/ABC-130k raw recordings,
Apache-2.0) in LeRobot v3.0 format.
This split: 129,032 episodes · 381,742,382 frames · 197 tasks · 30 fps.
Validation split: huzheyuan/abc130k_v3_val
(1,592 episodes / 189 tasks, task strings aligned to this split).
Why another port
The existing LeRobot port (lerobot/abc_130k_v3_train)… See the full description on the dataset page: https://huggingface.co/datasets/huzheyuan/abc130k_v3_train.ABC-130k
ABC-130k → FiftyOne (Native Multimodal MCAP)
Demo preview of ABC-130k subset in FiftyOne
Use FiftyOne to explore ABC-130k,
the largest open bimanual robot-teleoperation dataset (134,806 episodes / 195 tasks / 3,553 hours).
Each episode renders natively from its .mcap: synchronized multi-camera video with robot
arm/end-effector telemetry charted against the video timeline.
This dataset: Voxel51/ABC-130k — load directly with load_from_hub
Loader (FiftyOne):… See the full description on the dataset page: https://huggingface.co/datasets/Voxel51/ABC-130k.abc_130k_v3_valThis dataset was created using LeRobot.
Dataset Structure
meta/info.json:
{
"codebase_version": "v3.0",
"fps": 30,
"features": {
"observation.state": {
"dtype": "float32",
"shape": [
14
],
"names": [
"left_arm_joint_1",
"left_arm_joint_2",
"left_arm_joint_3",
"left_arm_joint_4",
"left_arm_joint_5"… See the full description on the dataset page: https://huggingface.co/datasets/lerobot/abc_130k_v3_val.abc_130k_trainThis dataset was created using LeRobot.
Dataset Structure
meta/info.json:
{
"codebase_version": "v3.0",
"fps": 30,
"features": {
"observation.images.top": {
"dtype": "video",
"shape": [
224,
224,
3
],
"names": [
"height",
"width",
"channels"
],
"info": {
"video.height":… See the full description on the dataset page: https://huggingface.co/datasets/typoverflow/abc_130k_train.abc_130k_v3_smokeThis dataset was created using LeRobot.
Dataset Structure
meta/info.json:
{
"codebase_version": "v3.0",
"fps": 30,
"features": {
"observation.state": {
"dtype": "float32",
"shape": [
14
],
"names": [
"left_arm_joint_1",
"left_arm_joint_2",
"left_arm_joint_3",
"left_arm_joint_4",
"left_arm_joint_5"… See the full description on the dataset page: https://huggingface.co/datasets/lerobot/abc_130k_v3_smoke.ABC-130k
ABC-130k — LeRobot v3.0 (train + val)
Bimanual teleoperation data from ABC-130k, converted to the
LeRobot v3.0 format. Robot type yam_bimanual, recorded at 10 fps with
three cameras. Upstream release and training code: xdof/ABC-130k
and amazon-far/abc.
Contents
Split
Subsets
Episodes
Frames
Size
train
459
79,047
66,540,229
4.43 TB
val
269
985
904,992
57 GB
Total
728
80,032
67,445,221
4,493,296,542,111 bytes
197 distinct tasks. A task appears… See the full description on the dataset page: https://huggingface.co/datasets/Dario-Shit2/ABC-130k.XDOF_ABC-130k_pantheon_annotated
Pantheon
Source dataset: XDOF/ABC-130k
This dataset is Pantheon's derived flags and annotations for the XDOF ABC-130k dataset. Reference our report https://pantheon.inc/research/we-looked-at-the-data for information about our methodology and issues we found.
See ./{version}/README.md (currently v1) for each version's annotation schema. You will need to merge the annotations with the original dataset. QC flags and processed actions may contain errors, false positives, and false… See the full description on the dataset page: https://huggingface.co/datasets/sambhav-pantheon/XDOF_ABC-130k_pantheon_annotated.abc130k_v3_val
ABC-130k — 256² full-frame re-conversion (validation split)
Validation split of the full-resolution ABC-130k re-conversion. 1,592 episodes ·
5,069,390 frames · 189 tasks · 30 fps (the tasks are a subset of the 197 train
tasks; task strings match the train split exactly).
See huzheyuan/abc130k_v3_train
for the full documentation: why this port exists (full-frame 256×256 squash vs the
letterboxed 224×224 LeRobot port), feature schema and conventions, and provenance.
Source:… See the full description on the dataset page: https://huggingface.co/datasets/huzheyuan/abc130k_v3_val.ABC-130k-val
ABC-130k — validation split
Bimanual teleoperation data in LeRobot v3.0 format, robot type yam_bimanual,
recorded at 10 fps.
269 task subsets, one directory each
985 episodes in total
~58 GB
Each subset is a self-contained LeRobot dataset:
<task_name>/
data/ # parquet episode data
videos/ # per-camera mp4
meta/ # info.json, tasks.parquet, stats.json, episode_metadata.parquet, ...
README.md
Directory names encode the task and the camera resolution, e.g.… See the full description on the dataset page: https://huggingface.co/datasets/Dario-Shit0/ABC-130k-val.abc_130k_valThis dataset was created using LeRobot.
Dataset Structure
meta/info.json:
{
"codebase_version": "v3.0",
"fps": 30,
"features": {
"observation.images.top": {
"dtype": "video",
"shape": [
224,
224,
3
],
"names": [
"height",
"width",
"channels"
],
"info": {
"video.height": 224… See the full description on the dataset page: https://huggingface.co/datasets/typoverflow/abc_130k_val.abc130k-screwdriver-mcap-lerobot-v3
ABC-130k MCAP → validated LeRobot v3 telemetry
Before → after: raw MCAP telemetry becomes a documented LeRobot v3.0 episode with preserved state/action mappings, provenance, and reader validation. Convert your MCAP free →
Community conversion produced by ViaCatalyst BYOD. This repository is not an official upstream release and is not affiliated with ABC-130k or Voxel51.
This is a provenance-complete conversion of one pinned MCAP episode from Voxel51's Apache-2.0 repackaging… See the full description on the dataset page: https://huggingface.co/datasets/ViaCatalyst/abc130k-screwdriver-mcap-lerobot-v3.ABC-130k-lerobot
ABC-130k (LeRobot v3)
LeRobot v3-format conversion of XDOF/ABC-130k,
the bimanual YAM-station manipulation dataset, converted from the original MCAP logs.
Contents
343 self-contained LeRobot v3 datasets, one per task unit (large tasks are split into
__partNN chunks — a multi-dataset loader consumes the full list).
118,685 episodes / 199 tasks.
232,430,937 frames (~3,228 h at 20 Hz).
3 cameras: observation.images.top, observation.images.left_wrist… See the full description on the dataset page: https://huggingface.co/datasets/LinhanWang/ABC-130k-lerobot.abc_130k_v3_smoke
abc_130k_v3_smoke (TsFile)
Apache TsFile version of lerobot/abc_130k_v3_smoke.
Overview
This dataset is a LeRobot v3.0 bimanual dataset (a smoke-test slice of the ABC-130K corpus) recorded with a YAM bimanual arm.
Robot: Yam bimanual (yam_bimanual)
Scale: 85 episodes, 313,094 frames, 30 fps
Split: train
Cameras (not uploaded): top, left_wrist, right_wrist
Schema (TsFile structure)
The TsFile table is named abc_130k_v3_smoke.
Role
Columns… See the full description on the dataset page: https://huggingface.co/datasets/THULab/abc_130k_v3_smoke.