LeRobot/LIBERO_10-Cosmos3 Dataset Release
lerobot/LIBERO_10-Cosmos3
The LIBERO-10 suite in the layout of the Cosmos3 LIBERO recipe, converted from nvidia/LIBERO_LeRobot_v3 (libero_10). It contains 379 episodes, 101469 frames, 10 tasks, and is recorded at 20 FPS using LeRobot dataset v3.0. The action space uses float32[10] with delta position, rot6d, and gripper channels, while the wrist camera observation is renamed to image2. This dataset was created using LeRobot, and the Cosmos3 policies of LeRobot fine-tune on it.
This release provides a converted dataset containing 379 episodes and 10 tasks from the LIBERO-10 suite, adapted to the LeRobot v3.0 format with specific action space changes (rot6d) and observation renaming for use by Cosmos3 policies.
Source: LeRobot — Dataset repository updates · Read original article ↗
Article text · Original source · English
On this page
The LIBERO-10 suite in the layout of the Cosmos3 LIBERO recipe, converted from nvidia/LIBERO_LeRobot_v3 (libero_10). 379 episodes, 101469 frames, 10 tasks, 20 FPS, LeRobot dataset v3.0.
This dataset was created using LeRobot. The Cosmos3 policies of LeRobot fine-tune on it (see the LeRobot Cosmos3 documentation).
Changes from the source
action: float32[10][delta position (3), rot6d (6), gripper]with the channel namespos_0topos_2,rot6d_0torot6d_5andgripper_0, instead of the float32[7][delta position (3), axis angle (3), gripper]of the source. rot6d holds the first two columns of the rotation matrix of the axis angle, and the gripper (1 open, 0 closed) passes through. This is the action of the native Cosmos3 LIBERO recipe.- The wrist camera
observation.images.wrist_imageis renamedobservation.images.image2, which is the camera key of the LeRobot LIBERO environment. - The action statistics in
meta/stats.json(mean, std, min, max, q01, q99) are those of the native Cosmos3 LIBERO recipe. - The suites are merged in the order libero_10 with shifted episode, frame and task indices. The videos are byte copies of the source.
Dataset structure
{
"codebase_version": "v3.0",
"fps": 20,
"features": {
"observation.images.image": {
"dtype": "video",
"shape": [
256,
256,
3
],
"names": [
"height",
"width",
"channels"
],
"info": {
"video.height": 256,
"video.width": 256,
"video.codec": "av1",
"video.pix_fmt": "yuv420p",
"video.is_depth_map": false,
"video.fps": 20,
"video.channels": 3,
"has_audio": false
}
},
"observation.images.image2": {
"dtype": "video",
"shape": [
256,
256,
3
],
"names": [
"height",
"width",
"channels"
],
"info": {
"video.height": 256,
"video.width": 256,
"video.codec": "av1",
"video.pix_fmt": "yuv420p",
"video.is_depth_map": false,
"video.fps": 20,
"video.channels": 3,
"has_audio": false
}
},
"observation.state": {
"dtype": "float32",
"shape": [
8
],
"names": {
"motors": [
"pos_0",
"pos_1",
"pos_2",
"axisangle_0",
"axisangle_1",
"axisangle_2",
"gripper_0",
"gripper_1"
]
},
"fps": 20
},
"action": {
"dtype": "float32",
"shape": [
10
],
"names": {
"motors": [
"pos_0",
"pos_1",
"pos_2",
"rot6d_0",
"rot6d_1",
"rot6d_2",
"rot6d_3",
"rot6d_4",
"rot6d_5",
"gripper_0"
]
},
"fps": 20
},
"timestamp": {
"dtype": "float32",
"shape": [
1
],
"names": null
},
"frame_index": {
"dtype": "int64",
"shape": [
1
],
"names": null
},
"episode_index": {
"dtype": "int64",
"shape": [
1
],
"names": null
},
"index": {
"dtype": "int64",
"shape": [
1
],
"names": null
},
"task_index": {
"dtype": "int64",
"shape": [
1
],
"names": null
}
},
"total_episodes": 379,
"total_frames": 101469,
"total_tasks": 10,
"chunks_size": 1000,
"data_files_size_in_mb": 100,
"video_files_size_in_mb": 200,
"data_path": "data/chunk-{chunk_index:03d}/file-{file_index:03d}.parquet",
"video_path": "videos/{video_key}/chunk-{chunk_index:03d}/file-{file_index:03d}.mp4",
"robot_type": "franka",
"splits": {
"train": "0:379"
}
}
License
The source dataset and this conversion use OpenMDW 1.1.
- Downloads last month
- -
Collection including lerobot/LIBERO_10-Cosmos3
What the source reports
Publisher-reported claims, with original evidence. These results have not been independently verified by RoboSignal.
Reported numbers
Demonstrations / episodes
379 episodes
View original evidence
The LIBERO-10 suite in the layout of the Cosmos3 LIBERO recipe, converted from nvidia/LIBERO_LeRobot_v3 (libero_10). 379 episodes, 101469 frames, 10 tasks, 20 FPS, LeRobot dataset v3.0. This dataset was created using LeRobot . The Cosmos3 policies of LeRobot fine-tune on it (see the LeRobot Cosmos3 documentation). Changes from the source action : float32[10] [delta position (3), rot6d (6), gripper] with the channel n
Open source S2frames
101,469 frames
View original evidence
The LIBERO-10 suite in the layout of the Cosmos3 LIBERO recipe, converted from nvidia/LIBERO_LeRobot_v3 (libero_10). 379 episodes, 101469 frames, 10 tasks, 20 FPS, LeRobot dataset v3.0. This dataset was created using LeRobot . The Cosmos3 policies of LeRobot fine-tune on it (see the LeRobot Cosmos3 documentation). Changes from the source action : float32[10] [delta position (3), rot6d (6), gripper] with the channel n
Open source S2tasks
10
View original evidence
The LIBERO-10 suite in the layout of the Cosmos3 LIBERO recipe, converted from nvidia/LIBERO_LeRobot_v3 (libero_10). 379 episodes, 101469 frames, 10 tasks, 20 FPS, LeRobot dataset v3.0. This dataset was created using LeRobot . The Cosmos3 policies of LeRobot fine-tune on it (see the LeRobot Cosmos3 documentation). Changes from the source action : float32[10] [delta position (3), rot6d (6), gripper] with the channel n
Open source S2data_files_size_in_mb
100
View original evidence
pe" : "int64" , "shape" : [ 1 ] , "names" : null } , "index" : { "dtype" : "int64" , "shape" : [ 1 ] , "names" : null } , "task_index" : { "dtype" : "int64" , "shape" : [ 1 ] , "names" : null } } , "total_episodes" : 379 , "total_frames" : 101469 , "total_tasks" : 10 , "chunks_size" : 1000 , "data_files_size_in_mb" : 100 , "video_files_size_in_mb" : 200 , "data_path" : "data/chunk-{chunk_index:03d}/file-{file_index:0
Open source S8video_files_size_in_mb
200
View original evidence
pe" : "int64" , "shape" : [ 1 ] , "names" : null } , "index" : { "dtype" : "int64" , "shape" : [ 1 ] , "names" : null } , "task_index" : { "dtype" : "int64" , "shape" : [ 1 ] , "names" : null } } , "total_episodes" : 379 , "total_frames" : 101469 , "total_tasks" : 10 , "chunks_size" : 1000 , "data_files_size_in_mb" : 100 , "video_files_size_in_mb" : 200 , "data_path" : "data/chunk-{chunk_index:03d}/file-{file_index:0
Open source S8fps
20
View original evidence
The LIBERO-10 suite in the layout of the Cosmos3 LIBERO recipe, converted from nvidia/LIBERO_LeRobot_v3 (libero_10). 379 episodes, 101469 frames, 10 tasks, 20 FPS, LeRobot dataset v3.0. This dataset was created using LeRobot . The Cosmos3 policies of LeRobot fine-tune on it (see the LeRobot Cosmos3 documentation). Changes from the source action : float32[10] [delta position (3), rot6d (6), gripper] with the channel n
Open source S2chunk_size
1,000
View original evidence
pe" : "int64" , "shape" : [ 1 ] , "names" : null } , "index" : { "dtype" : "int64" , "shape" : [ 1 ] , "names" : null } , "task_index" : { "dtype" : "int64" , "shape" : [ 1 ] , "names" : null } } , "total_episodes" : 379 , "total_frames" : 101469 , "total_tasks" : 10 , "chunks_size" : 1000 , "data_files_size_in_mb" : 100 , "video_files_size_in_mb" : 200 , "data_path" : "data/chunk-{chunk_index:03d}/file-{file_index:0
Open source S8
Dataset details reported by the source
- License
- OpenMDW 1.1 Open source S9
- Version
- v3.0 Open source S4
- Modalities
- video Open source S5state Open source S6action Open source S7
- Embodiments
- franka Open source S9
- paper: not_reported
- weights: not_reported
- dataset: Reported pending Open source S9
- code: not_reported
What remains unknown
Source excerpts and review record
Automatically extracted; no manual editorial approval recorded.
3d}.parquet" , "video_path" : "videos/{video_key}/chunk-{chunk_index:03d}/file-{file_index:03d}.mp4" , "robot_type" : "franka" , "splits" : { "train" : "0:379" } } License The source dataset and this conversion use OpenMDW 1.1 . Downloads last month - Collection including lerobot/LIBERO_10-Cosmos3
Open source S9
.image" : { "dtype" : "video" , "shape" : [ 256 , 256 , 3 ] , "names" : [ "height" , "width" , "channels" ] , "info" : { "video.height" : 256 , "video.width" : 256 , "video.codec" : "av1" , "video.pix_fmt" : "yuv420p" , "video.is_depth_map" : false , "video.fps" : 20 , "video.channels" : 3 , "has_audio" : false } } , "observation.images.image2" : { "dtype" : "video" , "shape" : [ 256 , 256 , 3 ] , "names" : [ "height
Open source S5
" , "width" , "channels" ] , "info" : { "video.height" : 256 , "video.width" : 256 , "video.codec" : "av1" , "video.pix_fmt" : "yuv420p" , "video.is_depth_map" : false , "video.fps" : 20 , "video.channels" : 3 , "has_audio" : false } } , "observation.state" : { "dtype" : "float32" , "shape" : [ 8 ] , "names" : { "motors" : [ "pos_0" , "pos_1" , "pos_2" , "axisangle_0" , "axisangle_1" , "axisangle_2" , "gripper_0" , "
Open source S6
gripper_1" ] } , "fps" : 20 } , "action" : { "dtype" : "float32" , "shape" : [ 10 ] , "names" : { "motors" : [ "pos_0" , "pos_1" , "pos_2" , "rot6d_0" , "rot6d_1" , "rot6d_2" , "rot6d_3" , "rot6d_4" , "rot6d_5" , "gripper_0" ] } , "fps" : 20 } , "timestamp" : { "dtype" : "float32" , "shape" : [ 1 ] , "names" : null } , "frame_index" : { "dtype" : "int64" , "shape" : [ 1 ] , "names" : null } , "episode_index" : { "dty
Open source S7
The LIBERO-10 suite in the layout of the Cosmos3 LIBERO recipe, converted from nvidia/LIBERO_LeRobot_v3 (libero_10). 379 episodes, 101469 frames, 10 tasks, 20 FPS, LeRobot dataset v3.0. This dataset was created using LeRobot . The Cosmos3 policies of LeRobot fine-tune on it (see the LeRobot Cosmos3 documentation). Changes from the source action : float32[10] [delta position (3), rot6d (6), gripper] with the channel n
Open source S2
is the camera key of the LeRobot LIBERO environment. The action statistics in meta/stats.json (mean, std, min, max, q01, q99) are those of the native Cosmos3 LIBERO recipe. The suites are merged in the order libero_10 with shifted episode, frame and task indices. The videos are byte copies of the source. Dataset structure meta/info.json : { "codebase_version" : "v3.0" , "fps" : 20 , "features" : { "observation.images
Open source S4
pe" : "int64" , "shape" : [ 1 ] , "names" : null } , "index" : { "dtype" : "int64" , "shape" : [ 1 ] , "names" : null } , "task_index" : { "dtype" : "int64" , "shape" : [ 1 ] , "names" : null } } , "total_episodes" : 379 , "total_frames" : 101469 , "total_tasks" : 10 , "chunks_size" : 1000 , "data_files_size_in_mb" : 100 , "video_files_size_in_mb" : 200 , "data_path" : "data/chunk-{chunk_index:03d}/file-{file_index:0
Open source S8
Implications for data suppliers
RoboSignal interpretation and collection questions, not statements of buyer demand.
- Confirm the required data type and collection setting with the buyer; this source does not establish a complete collection specification.
- Compare the reported units and scope before using these quantities in a budget. Recording hours, sensor-hours and trajectories are different measures.
- Check the exact dataset version, license, modalities and task coverage against the intended use.
- Validate demand and acceptance criteria with a buyer before scaling. Publication, popularity and a research result do not establish a purchase commitment.
Source：LeRobot — Dataset repository updates · huggingface.co