datasets
Training and evaluation data, with the modality, task and licence stated up front. Listed live from the Hugging Face Hub.
xarm_lift_mediumThis dataset was created using LeRobot.
Dataset Structure
meta/info.json:
{
"codebase_version": "v2.0",
"robot_type": "unknown",
"total_episodes": 800,
"total_frames": 20000,
"total_tasks": 1,
"total_videos": 800,
"total_chunks": 1,
"chunks_size": 1000,
"fps": 15,
"splits": {
"train": "0:800"
},
"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/lerobot/xarm_lift_medium.xarm_push_mediumThis dataset was created using LeRobot.
Dataset Structure
meta/info.json:
{
"codebase_version": "v2.0",
"robot_type": "unknown",
"total_episodes": 800,
"total_frames": 20000,
"total_tasks": 1,
"total_videos": 800,
"total_chunks": 1,
"chunks_size": 1000,
"fps": 15,
"splits": {
"train": "0:800"
},
"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/lerobot/xarm_push_medium.xarm_lift_medium_replayThis dataset was created using LeRobot.
Dataset Structure
meta/info.json:
{
"codebase_version": "v2.0",
"robot_type": "unknown",
"total_episodes": 800,
"total_frames": 20000,
"total_tasks": 1,
"total_videos": 800,
"total_chunks": 1,
"chunks_size": 1000,
"fps": 15,
"splits": {
"train": "0:800"
},
"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/lerobot/xarm_lift_medium_replay.xarm_push_medium_replayThis dataset was created using LeRobot.
Dataset Structure
meta/info.json:
{
"codebase_version": "v2.0",
"robot_type": "unknown",
"total_episodes": 800,
"total_frames": 20000,
"total_tasks": 1,
"total_videos": 800,
"total_chunks": 1,
"chunks_size": 1000,
"fps": 15,
"splits": {
"train": "0:800"
},
"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/lerobot/xarm_push_medium_replay.libero-island-ablation-v12-sigreg-medium
V12 — Single head + SIGReg medium weight (λ=0.01)
Bridges V3 (λ_sig=0.1, sig-dominated) and V8 (λ_sig=0.00025, action-dominated).
Architecture / Training
Backbone: Qwen2.5-VL-3B-Instruct + LoRA (r=32, q/v_proj)
Single AttentiveLatentHead A + ResNet action head
Loss: 1.0·L1_action + 0.01·SIGReg(z_a, z_domain_current)
100K steps, bs=64, lr=1e-4 cosine, 10 epochs
Results — epoch_10
Setting
SR
AA
0.72
BB
0.72
C_A
0.12
C_B
0.24
AL_B… See the full description on the dataset page: https://huggingface.co/datasets/disentangled-vla/libero-island-ablation-v12-sigreg-medium.robolab-mgh-mustard-mediumThis dataset was created using LeRobot.
Dataset Description
Simulated manipulation demonstrations generated by our MimicGen reimplementation on the RoboLab
(Isaac Lab) benchmark. One cell of a generator x task x difficulty grid (MimicGen) / generator x source-count grid (PGDG).
What this is
Generator
MimicGen (our reimplementation, not the authors' code)
Task
mustard
Initial-pose randomization
medium — position 50%, yaw ±25°
Episodes
3045… See the full description on the dataset page: https://huggingface.co/datasets/DAVIAN-Robotics/robolab-mgh-mustard-medium.pickcube_medium_nocamperturb_500This dataset was created using LeRobot.
Dataset Structure
meta/info.json:
{
"codebase_version": "v2.1",
"robot_type": "panda",
"total_episodes": 501,
"total_frames": 71968,
"total_tasks": 1,
"total_videos": 1002,
"total_chunks": 1,
"chunks_size": 1000,
"fps": 50,
"splits": {
"train": "0:501"
},
"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/dureduck/pickcube_medium_nocamperturb_500.so101_mimic_mediumThis dataset was created using LeRobot.
Dataset Structure
meta/info.json:
{
"codebase_version": "v3.0",
"fps": 30,
"features": {
"action": {
"dtype": "float32",
"shape": [
6
],
"names": [
"Rotation.pos",
"Pitch.pos",
"Elbow.pos",
"Wrist_Pitch.pos",
"Wrist_Roll.pos",
"Jaw.pos"
]… See the full description on the dataset page: https://huggingface.co/datasets/xBerry/so101_mimic_medium.robolab-mg-banana-mediumThis dataset was created using LeRobot.
Dataset Description
Simulated manipulation demonstrations generated by our MimicGen reimplementation on the RoboLab
(Isaac Lab) benchmark. One of 12 cells in a generator x task x difficulty grid.
What this is
Generator
MimicGen (our reimplementation, not the authors' code)
Task
banana
Initial-pose randomization
medium — position 50%, yaw ±25°
Episodes
3039 (successes only)
Generation success rate
0.279… See the full description on the dataset page: https://huggingface.co/datasets/DAVIAN-Robotics/robolab-mg-banana-medium.metra-alpha-ma-medium-a05-n3000This dataset was created using LeRobot.
Dataset Description
Successful BananaInBowl demonstrations collected from an RFCL-trained SAC policy in
Isaac Lab (RoboLab), for behaviour-cloning research on strategy diversity.
Lane: metra-alpha — a sweep of the METRA intrinsic-reward scale alpha, with every
other axis held fixed (banana task, 50 demos, sf=0.5, z_dim=3, phi_space=full, z_unit=true).
Each dataset is one (difficulty level, alpha) cell. Difficulty here: medium.
Collection:… See the full description on the dataset page: https://huggingface.co/datasets/DAVIAN-Robotics/metra-alpha-ma-medium-a05-n3000.robolab-mgh-banana-mediumThis dataset was created using LeRobot.
Dataset Description
Simulated manipulation demonstrations generated by our MimicGen reimplementation on the RoboLab
(Isaac Lab) benchmark. One cell of a generator x task x difficulty grid (MimicGen) / generator x source-count grid (PGDG).
What this is
Generator
MimicGen (our reimplementation, not the authors' code)
Task
banana
Initial-pose randomization
medium — position 50%, yaw ±25°
Episodes
3021… See the full description on the dataset page: https://huggingface.co/datasets/DAVIAN-Robotics/robolab-mgh-banana-medium.so101_real_mediumThis dataset was created using LeRobot.
Dataset Structure
meta/info.json:
{
"codebase_version": "v3.0",
"fps": 30,
"features": {
"action": {
"dtype": "float32",
"shape": [
6
],
"names": [
"Rotation.pos",
"Pitch.pos",
"Elbow.pos",
"Wrist_Pitch.pos",
"Wrist_Roll.pos",
"Jaw.pos"
]… See the full description on the dataset page: https://huggingface.co/datasets/xBerry/so101_real_medium.RoboLab-MustardInRightBin-medium-capx-oracle-260812This dataset was created using LeRobot.
Dataset Description
Simulated manipulation demonstrations generated in RoboLab (Isaac Lab) by a deterministic waypoint oracle with privileged state access (no cameras used for control). Task: MustardInRightBin, difficulty level medium.
What this is
Part of a 3 task x 3 difficulty grid of oracle datasets. The oracle replays a fixed waypoint recipe solved with rest-cost IK (pyroki) and is rate-limited to match the arm-speed… See the full description on the dataset page: https://huggingface.co/datasets/DAVIAN-Robotics/RoboLab-MustardInRightBin-medium-capx-oracle-260812.RoboLab-FoodPacking1Cans-medium-capx-oracle-260812This dataset was created using LeRobot.
Dataset Description
Simulated manipulation demonstrations generated in RoboLab (Isaac Lab) by a deterministic waypoint oracle with privileged state access (no cameras used for control). Task: FoodPacking1Cans, difficulty level medium.
What this is
Part of a 3 task x 3 difficulty grid of oracle datasets. The oracle replays a fixed waypoint recipe solved with rest-cost IK (pyroki) and is rate-limited to match the arm-speed cap… See the full description on the dataset page: https://huggingface.co/datasets/DAVIAN-Robotics/RoboLab-FoodPacking1Cans-medium-capx-oracle-260812.pb-hard-balance-medium-v1
PB Hard Balance Medium V1
This dataset is part of the IMBench suite,
distributed in the LeRobot v3.0 format.
Overview
Episodes
10
Total frames
11381
FPS
100
Robot
Panda
Cameras
4 (rgb_ego_cam, rgb_exo_left_cam, rgb_exo_right_cam, rgb_gripper_cam)
Visualize
Browse episodes interactively in the LeRobot Visualizer:
→ https://lerobot-visualize-dataset.hf.space/imbench/pb-hard-balance-medium-v1/episode_0
Load with LeRobot… See the full description on the dataset page: https://huggingface.co/datasets/imbench/pb-hard-balance-medium-v1.metra-alpha-ma-medium-a005-n3000This dataset was created using LeRobot.
Dataset Description
Successful BananaInBowl demonstrations collected from an RFCL-trained SAC policy in
Isaac Lab (RoboLab), for behaviour-cloning research on strategy diversity.
Lane: metra-alpha — a sweep of the METRA intrinsic-reward scale alpha, with every
other axis held fixed (banana task, 50 demos, sf=0.5, z_dim=3, phi_space=full, z_unit=true).
Each dataset is one (difficulty level, alpha) cell. Difficulty here: medium.
Collection:… See the full description on the dataset page: https://huggingface.co/datasets/DAVIAN-Robotics/metra-alpha-ma-medium-a005-n3000.metra-alpha-ma-medium-a01-n3000This dataset was created using LeRobot.
Dataset Description
Successful BananaInBowl demonstrations collected from an RFCL-trained SAC policy in
Isaac Lab (RoboLab), for behaviour-cloning research on strategy diversity.
Lane: metra-alpha — a sweep of the METRA intrinsic-reward scale alpha, with every
other axis held fixed (banana task, 50 demos, sf=0.5, z_dim=3, phi_space=full, z_unit=true).
Each dataset is one (difficulty level, alpha) cell. Difficulty here: medium.
Collection:… See the full description on the dataset page: https://huggingface.co/datasets/DAVIAN-Robotics/metra-alpha-ma-medium-a01-n3000.robolab-mgh-food-mediumThis dataset was created using LeRobot.
Dataset Description
Simulated manipulation demonstrations generated by our MimicGen reimplementation on the RoboLab
(Isaac Lab) benchmark. One cell of a generator x task x difficulty grid (MimicGen) / generator x source-count grid (PGDG).
What this is
Generator
MimicGen (our reimplementation, not the authors' code)
Task
food
Initial-pose randomization
medium — position 50%, yaw ±25°
Episodes
3060… See the full description on the dataset page: https://huggingface.co/datasets/DAVIAN-Robotics/robolab-mgh-food-medium.RoboLab-BananaInBowl-medium-capx-oracle-260809This dataset was created using LeRobot.
Dataset Description
Banana-in-bowl manipulation rollouts collected in RoboLab (Isaac Lab) by a deterministic scripted oracle, at the medium initial-pose difficulty level.
How this data was made
Not a learned policy. A privileged-state waypoint-replay controller reads ground-truth object poses (no cameras) and replays an 8-waypoint grasp-and-place sequence, rate-limited to the same arm speed cap as our RFCL policies so… See the full description on the dataset page: https://huggingface.co/datasets/DAVIAN-Robotics/RoboLab-BananaInBowl-medium-capx-oracle-260809.metra-alpha-ma-medium-a02-n3000This dataset was created using LeRobot.
Dataset Description
Successful BananaInBowl demonstrations collected from an RFCL-trained SAC policy in
Isaac Lab (RoboLab), for behaviour-cloning research on strategy diversity.
Lane: metra-alpha — a sweep of the METRA intrinsic-reward scale alpha, with every
other axis held fixed (banana task, 50 demos, sf=0.5, z_dim=3, phi_space=full, z_unit=true).
Each dataset is one (difficulty level, alpha) cell. Difficulty here: medium.
Collection:… See the full description on the dataset page: https://huggingface.co/datasets/DAVIAN-Robotics/metra-alpha-ma-medium-a02-n3000.metra-alpha-ma-medium-a00-n3000This dataset was created using LeRobot.
Dataset Description
Successful BananaInBowl demonstrations collected from an RFCL-trained SAC policy in
Isaac Lab (RoboLab), for behaviour-cloning research on strategy diversity.
Lane: metra-alpha — a sweep of the METRA intrinsic-reward scale alpha, with every
other axis held fixed (banana task, 50 demos, sf=0.5, z_dim=3, phi_space=full, z_unit=true).
Each dataset is one (difficulty level, alpha) cell. Difficulty here: medium.
Collection:… See the full description on the dataset page: https://huggingface.co/datasets/DAVIAN-Robotics/metra-alpha-ma-medium-a00-n3000.LLaVA-Video-medium-swiftmikasa-robo-vla-medium-lerobot
mikasa-robo-vla-medium-lerobot
LeRobotDataset v3 datasets for the 5 "Medium"-horizon MIKASA-Robo-VLA memory envs. One
dataset per env, as a top-level snake_case subfolder in this repo (same layout as
mikasa-robo/mikasa-robo-vla-lerobot).
Env code, install instructions, and an integration example for these 5 envs on top of stock
mikasa_robo_suite live in
UncertaintyVLA/mikasa-robo-vla-medium-envs —
start there if you want to run these envs yourself, not just replay the recorded… See the full description on the dataset page: https://huggingface.co/datasets/UncertaintyVLA/mikasa-robo-vla-medium-lerobot.robolab-mg-mustard-mediumThis dataset was created using LeRobot.
Dataset Description
Simulated manipulation demonstrations generated by our MimicGen reimplementation on the RoboLab
(Isaac Lab) benchmark. One of 12 cells in a generator x task x difficulty grid.
What this is
Generator
MimicGen (our reimplementation, not the authors' code)
Task
mustard
Initial-pose randomization
medium — position 50%, yaw ±25°
Episodes
3043 (successes only)
Generation success rate
0.245… See the full description on the dataset page: https://huggingface.co/datasets/DAVIAN-Robotics/robolab-mg-mustard-medium.pickcube_medium_nocamperturb_10_evalThis dataset was created using LeRobot.
Dataset Structure
meta/info.json:
{
"codebase_version": "v2.1",
"robot_type": "panda",
"total_episodes": 10,
"total_frames": 1203,
"total_tasks": 1,
"total_videos": 20,
"total_chunks": 1,
"chunks_size": 1000,
"fps": 50,
"splits": {
"train": "0:10"
},
"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/dureduck/pickcube_medium_nocamperturb_10_eval.libero-island-ablation-v13-infonce-medium-ours-target
V13 — Single head + InfoNCE medium weight (λ=0.1, ours target)
Bridges V7 (λ_nce=1.0, nce-dominated) and V9 (λ_nce=0.008, action-dominated).
Architecture / Training
Backbone: Qwen2.5-VL-3B-Instruct + LoRA (r=32, q/v_proj)
Single AttentiveLatentHead A + ResNet action head
V-JEPA target: our stage1 ViT-L + LoRA, 4-quad attentive future
Loss: 1.0·L1_action + 0.1·InfoNCE(z_a, z_target)
100K steps, bs=64, lr=1e-4 cosine, 10 epochs
Results — epoch_10… See the full description on the dataset page: https://huggingface.co/datasets/disentangled-vla/libero-island-ablation-v13-infonce-medium-ours-target.robolab-mg-food-mediumThis dataset was created using LeRobot.
Dataset Description
Simulated manipulation demonstrations generated by our MimicGen reimplementation on the RoboLab
(Isaac Lab) benchmark. One of 12 cells in a generator x task x difficulty grid.
What this is
Generator
MimicGen (our reimplementation, not the authors' code)
Task
food
Initial-pose randomization
medium — position 50%, yaw ±25°
Episodes
3044 (successes only)
Generation success rate
0.350… See the full description on the dataset page: https://huggingface.co/datasets/DAVIAN-Robotics/robolab-mg-food-medium.xarm_push_mediumThis dataset was created using LeRobot.
Dataset Structure
meta/info.json:
{
"codebase_version": "v2.0",
"robot_type": "unknown",
"total_episodes": 800,
"total_frames": 20000,
"total_tasks": 1,
"total_videos": 800,
"total_chunks": 1,
"chunks_size": 1000,
"fps": 15,
"splits": {
"train": "0:800"
},
"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/FT-LLM-2026-RAMEN/xarm_push_medium.insert_random_shaft_random_medium_gearThis dataset was created using LeRobot.
Dataset Structure
meta/info.json:
{
"codebase_version": "v3.0",
"robot_type": "aloha",
"total_episodes": 300,
"total_frames": 150000,
"total_tasks": 1,
"chunks_size": 1000,
"data_files_size_in_mb": 100,
"video_files_size_in_mb": 200,
"fps": 50,
"splits": {
"train": "0:300"
},
"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/lyl472324464/insert_random_shaft_random_medium_gear.bus_table_mediumThis dataset was created using LeRobot.
Dataset Structure
meta/info.json:
{
"codebase_version": "v3.0",
"robot_type": "so101_follower",
"total_episodes": 10,
"total_frames": 26384,
"total_tasks": 1,
"chunks_size": 1000,
"data_files_size_in_mb": 100,
"video_files_size_in_mb": 500,
"fps": 30,
"splits": {
"train": "0:10"
},
"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/lbxa/bus_table_medium.