datasets
Training and evaluation data, with the modality, task and licence stated up front. Listed live from the Hugging Face Hub.
pi05_droid_activationso101-pi05-base-dataset
SO-101 / SO-100 unified manipulation dataset
A single LeRobot dataset built from 150 public SO-101 / SO-100 teleoperation
datasets collected from the Hugging Face Hub, plus 6 recorded by the author.
The sources were screened for action-space conformity, deduplicated, and
rewritten into one schema with a fixed three-camera layout.
github.com/jinnymo/so101-pi05-base — the
scripts that built this, stage by stage: the Hub crawl, the screening criteria, the
action-convention check… See the full description on the dataset page: https://huggingface.co/datasets/dongyoonkim/so101-pi05-base-dataset.pi-0.5-dataset-hdf5Task1: Pick up the salt box
Task2: Pick up the garbage roll
Task3: Pick up the apple and drop it at the plate
Task4: Stack Cube
Task5: Hang on the hook
trossen_pi0pi05-physical-av-bcThis dataset was created using LeRobot.
Dataset Structure
meta/info.json:
{
"codebase_version": "v3.0",
"fps": 10,
"features": {
"observation.images.front": {
"dtype": "image",
"shape": [
480,
640,
3
],
"names": [
"height",
"width",
"channel"
]
},
"observation.state": {… See the full description on the dataset page: https://huggingface.co/datasets/markmusic/pi05-physical-av-bc.gripper-cableroutingtest2_pi05pi05_tea_pick_cuplibero90-pi05-frs-noise-labels
LIBERO-90 flow-reversal noise labels for pi0.5 (FRS-style offline DSBC)
Per-frame latent-noise labels for the success-filtered LIBERO-90 demos
(Embodied-CoT/embodied_features_and_demos_libero),
computed by reverse-integrating the frozen
pi0.5-LIBERO flow ODE
(plain reverse Euler, 10 steps, mirroring openpi's forward discretization)
at each demo action chunk — the offline labeling step of Flow Reversal
Steering (FRS)-style distilled noise-space BC.
Files
One .npz… See the full description on the dataset page: https://huggingface.co/datasets/knightnemo/libero90-pi05-frs-noise-labels.eval_pi0_inference_only_datasetThis dataset was created using LeRobot.
Dataset Structure
meta/info.json:
{
"codebase_version": "v3.0",
"robot_type": "so101_follower",
"total_episodes": 70,
"total_frames": 40583,
"total_tasks": 1,
"chunks_size": 1000,
"data_files_size_in_mb": 100,
"video_files_size_in_mb": 500,
"fps": 30,
"splits": {
"train": "0:70"
},
"data_path": "data/chunk-{chunk_index:03d}/file-{file_index:03d}.parquet",
"video_path":… See the full description on the dataset page: https://huggingface.co/datasets/tersooawai/eval_pi0_inference_only_dataset.Spa_Bench_Full_Pi0.5
Spa-Bench physical rollouts — pi0.5
Physical SO-101 rollout recordings for the pi0.5 Spa-Bench checkpoint at epoch
12, step 76,596.
Scope and counts
The scored benchmark protocol is episodes 0–899: 300 primary withheld/OOD
spatial trials and 600 diagnostic/control trials. Their prompt sequence matches
the MolmoAct2 and GR00T-N1.7 Frozen LLM full releases exactly.
This repository contains 920 episodes, 781,538 frames, 356 unique instruction
strings, two 480×640 RGB… See the full description on the dataset page: https://huggingface.co/datasets/justintiensmith/Spa_Bench_Full_Pi0.5.route_subtasks_dual_overhead_pi05
Route Subtask Episodes with Dual Overhead Views
This is a LeRobot v2.1 training dataset at 100 Hz. Every output episode contains
exactly one semantic subtask. Route boundaries are inferred from maximal
contiguous runs of metadata.route.active_subgoal; partial source episodes are
kept as the suffix subtasks they contain.
The dataset has 523 split route episodes and 135699 frames.
Camera streams
observation.images.video_overhead_vector: overhead image with the… See the full description on the dataset page: https://huggingface.co/datasets/ajaysri/route_subtasks_dual_overhead_pi05.eval_pi0_pretrainedThis dataset was created using LeRobot.
Dataset Structure
meta/info.json:
{
"codebase_version": "v3.0",
"robot_type": "so101_follower",
"total_episodes": 15,
"total_frames": 9094,
"total_tasks": 1,
"chunks_size": 1000,
"data_files_size_in_mb": 100,
"video_files_size_in_mb": 500,
"fps": 30,
"splits": {
"train": "0:15"
},
"data_path": "data/chunk-{chunk_index:03d}/file-{file_index:03d}.parquet",
"video_path":… See the full description on the dataset page: https://huggingface.co/datasets/tersooawai/eval_pi0_pretrained.ur5e_pi05_rollouts
UR5e pi0.5 rollouts
Policy rollouts of mahgoobi/ur5e_pi05_10k on a
real UR5e with a Robotiq gripper, task "put the cup in the bowl", collected 2026-09-01 from
the deployment page in RoboResearch
(roboresearch.evaluation.ur5e.ui). 25 episodes at 20 Hz, action chunks of 50 steps
with action_steps of them executed per chunk (25 for every episode but the first, which
ran 10). 20 successes, 5 failures. Every failure is in a scene with
three or more distractors; the clean and… See the full description on the dataset page: https://huggingface.co/datasets/mahgoobi/ur5e_pi05_rollouts.robotwin2-pi05-calibration-clean
RoboTwin 2.0 pi0.5 Calibration Subset (clean)
Sampled from TianxingChen/RoboTwin2.0 on HuggingFace.
Embodiment: aloha-agilex
Setting: clean (no domain randomization)
Tasks: 50
Episodes per task: 2
Total episodes: 100
Sampling seed (numpy default_rng): 42
Layout
<task>/
scene_info.json # task scene metadata
episode{N}.hdf5 # sensor data (rgb / state / action)
episode{N}.pkl # trajectory metadata
episode{N}.json # per-episode instruction pool
See… See the full description on the dataset page: https://huggingface.co/datasets/JingwuLuo/robotwin2-pi05-calibration-clean.pi0_RANDOM_RACKpi0_conversion_no_padThis dataset was created using LeRobot.
Dataset Structure
meta/info.json:
{
"codebase_version": "v2.0",
"robot_type": "panda",
"total_episodes": 1417,
"total_frames": 166855,
"total_tasks": 33,
"total_videos": 0,
"total_chunks": 2,
"chunks_size": 1000,
"fps": 15,
"splits": {
"train": "0:1417"
},
"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/mlfu7/pi0_conversion_no_pad.RoboTwin-adjust_bottle-official-demo_clean50-Pi0_processed-datarobosuite-lift-pi05
Robosuite Lift — π0.5 zero-shot rollouts (1000 episodes)
1000 rollouts of the π0.5 policy (pi05_libero checkpoint) operating zero-shot on the
robosuite Lift task ("lift the red cube").
No fine-tuning was done on Lift — the LIBERO checkpoint is applied directly, since LIBERO is built
on robosuite and exposes an identical Panda + OSC_POSE interface.
Every rollout is kept, success or failure — each episode carries a success flag. Failures are
intentionally included (they are useful… See the full description on the dataset page: https://huggingface.co/datasets/jacob3333/robosuite-lift-pi05.pi0fast-metaworld-activations-v1-ml45train-16envpi0_conversionThis dataset was created using LeRobot.
Dataset Structure
meta/info.json:
{
"codebase_version": "v2.0",
"robot_type": "panda",
"total_episodes": 1417,
"total_frames": 166855,
"total_tasks": 33,
"total_videos": 0,
"total_chunks": 2,
"chunks_size": 1000,
"fps": 15,
"splits": {
"train": "0:1417"
},
"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/mlfu7/pi0_conversion.behavior-pi05-all100-rollouts
BEHAVIOR pi0.5 all-100 rollouts
Public storage for videos and per-episode metrics from the all-100 experiment. Evaluation and replay run on the existing Linux machine, not Nebius.
No trained all-100 policy evaluation has been published yet. Every result must identify its policy commit, task, instance, seed, success, Q-score, steps and termination reason. Training replay and evaluation episodes remain separate.
Experiment plan and policies: behavior-pi05-all100.
eval_pi0_datasetThis dataset was created using LeRobot.
Dataset Structure
meta/info.json:
{
"codebase_version": "v3.0",
"robot_type": "so101_follower",
"total_episodes": 30,
"total_frames": 16013,
"total_tasks": 1,
"chunks_size": 1000,
"data_files_size_in_mb": 100,
"video_files_size_in_mb": 500,
"fps": 30,
"splits": {
"train": "0:30"
},
"data_path": "data/chunk-{chunk_index:03d}/file-{file_index:03d}.parquet",
"video_path":… See the full description on the dataset page: https://huggingface.co/datasets/tersooawai/eval_pi0_dataset.eval_pi0_bowl_plate_d435i_5ep_depth_datasetThis dataset was created using LeRobot.
Dataset Structure
meta/info.json:
{
"codebase_version": "v3.0",
"robot_type": "so101_follower",
"total_episodes": 75,
"total_frames": 44465,
"total_tasks": 1,
"chunks_size": 1000,
"data_files_size_in_mb": 100,
"video_files_size_in_mb": 500,
"fps": 30,
"splits": {
"train": "0:75"
},
"data_path": "data/chunk-{chunk_index:03d}/file-{file_index:03d}.parquet",
"video_path":… See the full description on the dataset page: https://huggingface.co/datasets/tersooawai/eval_pi0_bowl_plate_d435i_5ep_depth_dataset.libero-spatial-pi0-pi0This dataset was created using LeRobot.
Dataset Structure
meta/info.json:
{
"codebase_version": "v2.1",
"robot_type": "panda",
"total_episodes": 432,
"total_frames": 52970,
"total_tasks": 10,
"total_videos": 0,
"total_chunks": 1,
"chunks_size": 1000,
"fps": 10,
"splits": {
"train": "0:432"
},
"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/k1000dai/libero-spatial-pi0-pi0.route_plus_steer_place_subtasks_dual_overhead_pi05_224_plus_dagger
224px Route + Steering Placement + DAgger with Dual Overhead Views
This is a LeRobot v2.1 training dataset at 100 Hz. Every output episode contains
one constant prompt. Base route boundaries are inferred from maximal
contiguous runs of metadata.route.active_subgoal; partial source episodes are
kept as the suffix subtasks they contain.
The base data contains 495 route subtask episodes (128742 frames) and 202 steering/placement episodes (50099 frames). All 66 route DAgger clips… See the full description on the dataset page: https://huggingface.co/datasets/ajaysri/route_plus_steer_place_subtasks_dual_overhead_pi05_224_plus_dagger.eval_pi0_3D_infe_white_bowlThis dataset was created using LeRobot.
Dataset Structure
meta/info.json:
{
"codebase_version": "v3.0",
"robot_type": "so101_follower",
"total_episodes": 15,
"total_frames": 8315,
"total_tasks": 1,
"chunks_size": 1000,
"data_files_size_in_mb": 100,
"video_files_size_in_mb": 500,
"fps": 30,
"splits": {
"train": "0:15"
},
"data_path": "data/chunk-{chunk_index:03d}/file-{file_index:03d}.parquet",
"video_path":… See the full description on the dataset page: https://huggingface.co/datasets/tersooawai/eval_pi0_3D_infe_white_bowl.pi05-task3-demosmaniskill_pi0This dataset was created using LeRobot.
Dataset Structure
meta/info.json:
{
"codebase_version": "v2.0",
"robot_type": "panda",
"total_episodes": 200,
"total_frames": 21321,
"total_tasks": 1,
"total_videos": 0,
"total_chunks": 1,
"chunks_size": 1000,
"fps": 50,
"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/Ruoxiang/maniskill_pi0.route_subtasks_dual_overhead_pi05_448
Route Subtask Episodes with Dual Overhead Views
This is a LeRobot v2.1 training dataset at 100 Hz. Every output episode contains
one constant prompt. Base route boundaries are inferred from maximal
contiguous runs of metadata.route.active_subgoal; partial source episodes are
kept as the suffix subtasks they contain.
The dataset has 495 split route episodes and 128742 frames.
Camera streams
observation.images.video_overhead_vector: overhead image with the fixed… See the full description on the dataset page: https://huggingface.co/datasets/ajaysri/route_subtasks_dual_overhead_pi05_448.route_plus_steer_place_subtasks_dual_overhead_pi05
Route + Steering Placement Subtask Episodes with Dual Overhead Views
This is a LeRobot v2.1 training dataset at 100 Hz. Every output episode contains
exactly one semantic subtask. Route boundaries are inferred from maximal
contiguous runs of metadata.route.active_subgoal; partial source episodes are
kept as the suffix subtasks they contain.
The route portion has 495 split episodes and 128742 frames. The steering supplement contributes 202 episodes and 50099 frames without… See the full description on the dataset page: https://huggingface.co/datasets/ajaysri/route_plus_steer_place_subtasks_dual_overhead_pi05.