actions list | primary_images_jpeg listlengths 15.4k 23.7k | proprio list | wrist_images_jpeg listlengths 4.05k 25.9k |
|---|---|---|---|
[
0.053571428571428575,
-0.3883928571428572,
0,
0.013928571428571429,
0.06964285714285715,
-0.0064285714285714285,
-1
] | [
255,
216,
255,
224,
0,
16,
74,
70,
73,
70,
0,
1,
1,
0,
0,
1,
0,
1,
0,
0,
255,
219,
0,
67,
0,
2,
1,
1,
1,
1,
1,
2,
1,
1,
1,
2,
2,
2,
2,
2,
4,
3,
2,
2,
2,
2,
5,
4,
4,
3,
4,
6,
5,
6,
6,
6,
5,
6,
6,
6,
... | [
0.0387905347075188,
-0.03878512142606915,
-0.06444263097187469,
-0.023880958865721703,
0.6766150664025833,
0.9995150507267224,
0.0002630199214511131,
-0.031133784390390516,
-0.0005307172593940254
] | [
255,
216,
255,
224,
0,
16,
74,
70,
73,
70,
0,
1,
1,
0,
0,
1,
0,
1,
0,
0,
255,
219,
0,
67,
0,
2,
1,
1,
1,
1,
1,
2,
1,
1,
1,
2,
2,
2,
2,
2,
4,
3,
2,
2,
2,
2,
5,
4,
4,
3,
4,
6,
5,
6,
6,
6,
5,
6,
6,
6,
... |
[0.05625,-0.3803571428571429,-0.0,0.012857142857142857,0.06857142857142857,-0.0064285714285714285,-1(...TRUNCATED) | [255,216,255,224,0,16,74,70,73,70,0,1,1,0,0,1,0,1,0,0,255,219,0,67,0,2,1,1,1,1,1,2,1,1,1,2,2,2,2,2,4(...TRUNCATED) | [0.038851427314573826,-0.038937677004794086,-0.06518219281907443,-0.02589050966124446,0.675849534053(...TRUNCATED) | [255,216,255,224,0,16,74,70,73,70,0,1,1,0,0,1,0,1,0,0,255,219,0,67,0,2,1,1,1,1,1,2,1,1,1,2,2,2,2,2,4(...TRUNCATED) |
[0.05089285714285714,-0.39642857142857146,-0.0,0.015000000000000001,0.06857142857142857,-0.009642857(...TRUNCATED) | [255,216,255,224,0,16,74,70,73,70,0,1,1,0,0,1,0,1,0,0,255,219,0,67,0,2,1,1,1,1,1,2,1,1,1,2,2,2,2,2,4(...TRUNCATED) | [0.03889761782693676,-0.03900351944480077,-0.0659747241285795,-0.02992530020021744,0.674796970863628(...TRUNCATED) | [255,216,255,224,0,16,74,70,73,70,0,1,1,0,0,1,0,1,0,0,255,219,0,67,0,2,1,1,1,1,1,2,1,1,1,2,2,2,2,2,4(...TRUNCATED) |
[0.045535714285714284,-0.4205357142857143,-0.0,0.018214285714285714,0.07071428571428573,-0.011785714(...TRUNCATED) | [255,216,255,224,0,16,74,70,73,70,0,1,1,0,0,1,0,1,0,0,255,219,0,67,0,2,1,1,1,1,1,2,1,1,1,2,2,2,2,2,4(...TRUNCATED) | [0.038942586202864164,-0.03904874465482112,-0.06652170354247534,-0.03457109355933155,0.6738912205121(...TRUNCATED) | [255,216,255,224,0,16,74,70,73,70,0,1,1,0,0,1,0,1,0,0,255,219,0,67,0,2,1,1,1,1,1,2,1,1,1,2,2,2,2,2,4(...TRUNCATED) |
[0.03482142857142857,-0.45267857142857143,-0.0,0.023571428571428573,0.0675,-0.015000000000000001,-1.(...TRUNCATED) | [255,216,255,224,0,16,74,70,73,70,0,1,1,0,0,1,0,1,0,0,255,219,0,67,0,2,1,1,1,1,1,2,1,1,1,2,2,2,2,2,4(...TRUNCATED) | [0.03898561229430962,-0.039091816563277454,-0.06692780895195195,-0.0396323653744267,0.67313184220303(...TRUNCATED) | [255,216,255,224,0,16,74,70,73,70,0,1,1,0,0,1,0,1,0,0,255,219,0,67,0,2,1,1,1,1,1,2,1,1,1,2,2,2,2,2,4(...TRUNCATED) |
[0.048214285714285716,-0.5116071428571428,0.013392857142857144,0.03214285714285715,0.073928571428571(...TRUNCATED) | [255,216,255,224,0,16,74,70,73,70,0,1,1,0,0,1,0,1,0,0,255,219,0,67,0,2,1,1,1,1,1,2,1,1,1,2,2,2,2,2,4(...TRUNCATED) | [0.03902656923753971,-0.039133261523623064,-0.06723315687922576,-0.045092885138901743,0.672526166329(...TRUNCATED) | [255,216,255,224,0,16,74,70,73,70,0,1,1,0,0,1,0,1,0,0,255,219,0,67,0,2,1,1,1,1,1,2,1,1,1,2,2,2,2,2,4(...TRUNCATED) |
[0.04285714285714286,-0.5651785714285715,0.026785714285714288,0.04071428571428572,0.0739285714285714(...TRUNCATED) | [255,216,255,224,0,16,74,70,73,70,0,1,1,0,0,1,0,1,0,0,255,219,0,67,0,2,1,1,1,1,1,2,1,1,1,2,2,2,2,2,4(...TRUNCATED) | [0.039065118834812854,-0.03917357745722336,-0.06746739031758248,-0.05109637217445458,0.6720561079628(...TRUNCATED) | [255,216,255,224,0,16,74,70,73,70,0,1,1,0,0,1,0,1,0,0,255,219,0,67,0,2,1,1,1,1,1,2,1,1,1,2,2,2,2,2,4(...TRUNCATED) |
[0.07232142857142856,-0.6214285714285714,0.02142857142857143,0.047142857142857146,0.0728571428571428(...TRUNCATED) | [255,216,255,224,0,16,74,70,73,70,0,1,1,0,0,1,0,1,0,0,255,219,0,67,0,2,1,1,1,1,1,2,1,1,1,2,2,2,2,2,4(...TRUNCATED) | [0.03910154103259267,-0.03921263521887413,-0.0676630709334205,-0.05775578861604919,0.671759831937098(...TRUNCATED) | [255,216,255,224,0,16,74,70,73,70,0,1,1,0,0,1,0,1,0,0,255,219,0,67,0,2,1,1,1,1,1,2,1,1,1,2,2,2,2,2,4(...TRUNCATED) |
[0.1125,-0.6669642857142857,-0.0,0.05142857142857143,0.056785714285714294,-0.016071428571428573,-1.0(...TRUNCATED) | [255,216,255,224,0,16,74,70,73,70,0,1,1,0,0,1,0,1,0,0,255,219,0,67,0,2,1,1,1,1,1,2,1,1,1,2,2,2,2,2,4(...TRUNCATED) | [0.039135844962309665,-0.0392505711007094,-0.06768850964635308,-0.06511896454450018,0.67155615914174(...TRUNCATED) | [255,216,255,224,0,16,74,70,73,70,0,1,1,0,0,1,0,1,0,0,255,219,0,67,0,2,1,1,1,1,1,2,1,1,1,2,2,2,2,2,4(...TRUNCATED) |
[
0.13660714285714284,
-0.6883928571428571,
0,
0.05464285714285714,
0.045,
-0.016071428571428573,
-1
] | [255,216,255,224,0,16,74,70,73,70,0,1,1,0,0,1,0,1,0,0,255,219,0,67,0,2,1,1,1,1,1,2,1,1,1,2,2,2,2,2,4(...TRUNCATED) | [0.0391683008543409,-0.03928725411559123,-0.06723368789686796,-0.07317509822158301,0.671280700210511(...TRUNCATED) | [255,216,255,224,0,16,74,70,73,70,0,1,1,0,0,1,0,1,0,0,255,219,0,67,0,2,1,1,1,1,1,2,1,1,1,2,2,2,2,2,4(...TRUNCATED) |
nvidia/LIBERO-Cosmos-Policy(複製)
データの内容
LIBERO のロボット操作のシミュレーションベンチマークを改変したデータ。Cosmos Policy(Cosmos-Policy-LIBERO-Predict2-2B)の学習に使われ、4 つのタスクスイートを収める。HDF5 には、第三者視点と手首のカメラの JPEG 画像、関節とエンドエフェクタの状態、行動が入る。元の LIBERO を、解像度を 128 から 256 に上げ、no-op の行動を除き、決定的なシードで再生し直して作った。success_only は成功したデモだけ、all_episodes は成功と失敗の両方を含む。
データの概要
| 項目 | 値 |
|---|---|
| 行数 | 約 642,519(1 構成・1 分割の合計) |
| 列数 | 4 |
| 容量 | 27.57 GB(ファイル 2050 本) |
| ファイルの形式 | hdf5 198 |
| 公開 | public |
構成と分割
| 構成(config) | 分割(split)ごとの行数 | 列数 |
|---|---|---|
default |
train 約 642,519 |
4 |
列(構成 default・分割 train)
| 列 | 型 | 先頭の行の値 |
|---|---|---|
actions |
list<float64> |
[7 件] |
primary_images_jpeg |
list<uint8> |
[20164 件] |
proprio |
list<float64> |
[9 件] |
wrist_images_jpeg |
list<uint8> |
[18506 件] |
行数と列は HF のデータセットビューアの集計(元のデータセットの現在の版で数えたもの)。全データセットの一覧は HYUKI5/dataset-catalog にある。
複製について
nvidia/LIBERO-Cosmos-Policy をリビジョン b463cbd5d114909e4daea65eb03bed9787290042 の時点でそのまま複製したもの。中身は変えていない。判断モデル Laya(convaiinnovations/laya)の微調整に使う候補として、調査メモ(Laya 学習用データセット調査)に載せたデータを集めている。
出典
- 元のデータセット: https://huggingface.co/datasets/nvidia/LIBERO-Cosmos-Policy
- 複製したリビジョン:
b463cbd5d114909e4daea65eb03bed9787290042(2026-02-05 05:47:07+00:00) - 元のライセンス表記:
cc
論文
アノテーションの方法
元のカードにアノテーションの節が無かった。下の「調査メモの記述」と「元のカード」を参照。
微調整に使うときの注意
- 調査メモの台帳での区分: D(記載なし・other・unknown・資料間の食い違い)。区分は記載の文言で機械的に付けたもので、法的な判定ではない。学習に入れる前に元のカードとライセンスの原文で確かめる。
- Laya の入力は 2,048 token まで(ModernBERT-large / mmBERT)。長い本文は先頭から黙って切られるので、切れる行は事前に除くか、要約・切り出しをしてから入れる。
- 分割は記事・文書単位で行い、同じ本文から作った別の問いを学習側と較正側に分けない。温度と閾値は学習に使っていない分割で別々に当てる。
調査メモの記述
| メモ | 節 | ライセンス(記載) | 確認 | 注記 |
|---|---|---|---|---|
nvidia_nemotron.md |
付録. 取得した nvidia 組織のデータセット全件の簡易一覧 (L477) | cc | 確認済 | 確認印なし(各ファイルの凡例により確認済扱い) / 原文注記: × 非テキスト(ロボット・自動運転・3D・音声・画像・科学計算)または翻訳版ベンチ |
元のカード(日本語訳)
以下は元のデータセットのカードの本文を日本語に訳したもの(機械翻訳)。訳に疑問があるときは、末尾の原文を確かめる。
LIBERO-Cosmos-Policy
データセットの説明
LIBERO-Cosmos-Policy は、LIBERO シミュレーションベンチマークのデータセット を改変したもので、Cosmos Policy プロジェクトの一部として作られた。これは Cosmos-Policy-LIBERO-Predict2-2B チェックポイントの訓練に使われたデータセットである。
主な改変点
改変の内容は次のとおりである。
- 高解像度の画像: 画像は 256×256 ピクセルで保存される(元は 128×128)。
- no-op 行動の除去: ロボットの状態を変えない「no-op」(ゼロ)行動の遷移を取り除いた。
- JPEG 圧縮: ストレージの要件を減らすため、画像は JPEG で圧縮されている。
- 決定的な再生成: すべてのデモンストレーションを、再現性のために決定的なシードでシミュレーション環境に再生した。
データセットの構造
データセットは 2 つの主要なディレクトリに分かれる。
success_only/: 成功したデモンストレーションのエピソードだけを含む(フィルタ済みの版)。シミュレーション環境で再生したときに成功したデモンストレーションである(一部のデモンストレーションは遠隔操作中の人為的ミスで失敗することがわかっている)。このセットは、高品質な行動を生成するよう Cosmos Policy を訓練するために使われる。all_episodes/: 成功と失敗の両方を含むすべてのエピソード。このセットは、Cosmos Policy の世界モデルと価値関数の訓練に使われる。
各ディレクトリには、4 つのタスクスイートのデータが含まれる。LIBERO-Spatial、LIBERO-Object、LIBERO-Goal、LIBERO-10(LIBERO-Long とも呼ばれる)である。
libero_spatial- 空間推論を試すタスク(10 タスク、各 50 件のデモ)libero_object- 物体操作を試すタスク(10 タスク、各 50 件のデモ)libero_goal- 言語で指定された目標を試すタスク(10 タスク、各 50 件のデモ)libero_10- 長い手順の操作タスク(10 タスク、各 50 件のデモ)
データの形式
success_only/ の各 HDF5 ファイルには次が含まれる。
data/
├── demo_0/
│ ├── obs/
│ │ ├── agentview_rgb_jpeg # Third-person camera images (JPEG compressed)
│ │ ├── eye_in_hand_rgb_jpeg # Wrist camera images (JPEG compressed)
│ │ ├── gripper_states # Gripper joint positions
│ │ ├── joint_states # Robot joint positions
│ │ ├── ee_states # End-effector states (position + orientation)
│ │ ├── ee_pos # End-effector position
│ │ └── ee_ori # End-effector orientation
│ ├── actions # Action sequence
│ ├── states # Environment states
│ ├── robot_states # Combined robot state (gripper + EEF pos + EEF quat)
│ ├── rewards # Sparse rewards (0 until final timestep)
│ └── dones # Episode termination flags
├── demo_1/
...
all_episodes/ ディレクトリには、別の形式でロールアウトのデータが入っている。各エピソードは、次の命名パターンの別々の HDF5 ファイルとして保存される。
episode_data--suite={suite_name}--{timestamp}--task={task_id}--ep={episode_num}--success={True/False}--regen_demo.hdf5
これらの HDF5 ファイルにはそれぞれ次が含まれる。
# Datasets (arrays)
primary_images_jpeg # Third-person camera images (JPEG compressed), shape: (T, H, W, 3)
wrist_images_jpeg # Wrist camera images (JPEG compressed), shape: (T, H, W, 3)
proprio # Proprioceptive state (gripper + EEF pos + quat), shape: (T, 9)
actions # Action sequence, shape: (T, 7)
# Attributes (scalars/metadata)
success # Boolean: True if episode succeeded, False otherwise
task_description # String: Natural language task description
統計
- デモンストレーションの総数: フィルタの前は、タスクスイートごとに 500 件(4 スイートで計 2000 件)
- 成功率: 約 80〜90%(タスクスイートによって異なる)
- 画像の解像度: 256×256×3(RGB)
- 行動の次元: 7(6 自由度のエンドエフェクタ制御とグリッパー 1)
- 固有受容状態(proprioception)の次元: 9(グリッパーの関節 2、エンドエフェクタの位置 3、クォータニオン 4)
元の LIBERO データセット
このデータセットは、元の LIBERO ベンチマークから派生している。
- 論文: LIBERO: Benchmarking Knowledge Transfer for Lifelong Robot Learning
- リポジトリ: https://github.com/Lifelong-Robot-Learning/LIBERO
- ライセンス: CC BY 4.0
引用
このデータセットを使う場合は、元の LIBERO の論文と Cosmos Policy の論文の両方を引用してほしい。
ライセンス
Creative Commons Attribution 4.0 International(CC BY 4.0)
原文(英語)
LIBERO-Cosmos-Policy
Dataset Description
LIBERO-Cosmos-Policy is a modified version of the LIBERO simulation benchmark dataset, created as part of the Cosmos Policy project. This is the dataset used to train the Cosmos-Policy-LIBERO-Predict2-2B checkpoint.
Key Modifications
Our modifications include the following:
- Higher-resolution images: Images are saved at 256×256 pixels (vs. 128×128 in the original).
- No-op actions filtering: Transitions with "no-op" (zero) actions that don't change the robot's state are filtered out.
- JPEG compression: Images are JPEG-compressed to reduce storage requirements.
- Deterministic regeneration: All demonstrations are replayed in the simulation environment with deterministic seeding for reproducibility.
Dataset Structure
The dataset is organized into two main directories:
success_only/: Contains only successful demonstration episodes (filtered version). These are demonstrations that succeeded when replayed in the simulation environments (it turns out that some demonstrations fail due to human errors during teleoperation). This set is used to train Cosmos Policy to generate high-quality actions.all_episodes/: Contains all episodes, including both successful and failed demonstrations. This set is used to train Cosmos Policy's world model and value function.
Each directory contains data from four task suites: LIBERO-Spatial, LIBERO-Object, LIBERO-Goal, and LIBERO-10 (also called LIBERO-Long).
libero_spatial- Tasks testing spatial reasoning (10 tasks, 50 demos each)libero_object- Tasks testing object manipulation (10 tasks, 50 demos each)libero_goal- Tasks testing language-specified goals (10 tasks, 50 demos each)libero_10- Long-horizon manipulation tasks (10 tasks, 50 demos each)
Data Format
Each HDF5 file in success_only/ contains:
data/
├── demo_0/
│ ├── obs/
│ │ ├── agentview_rgb_jpeg # Third-person camera images (JPEG compressed)
│ │ ├── eye_in_hand_rgb_jpeg # Wrist camera images (JPEG compressed)
│ │ ├── gripper_states # Gripper joint positions
│ │ ├── joint_states # Robot joint positions
│ │ ├── ee_states # End-effector states (position + orientation)
│ │ ├── ee_pos # End-effector position
│ │ └── ee_ori # End-effector orientation
│ ├── actions # Action sequence
│ ├── states # Environment states
│ ├── robot_states # Combined robot state (gripper + EEF pos + EEF quat)
│ ├── rewards # Sparse rewards (0 until final timestep)
│ └── dones # Episode termination flags
├── demo_1/
...
The all_episodes/ directory contains rollout data in a different format. Each episode is stored as a separate HDF5 file with the naming pattern:
episode_data--suite={suite_name}--{timestamp}--task={task_id}--ep={episode_num}--success={True/False}--regen_demo.hdf5
Each of these HDF5 files contains:
# Datasets (arrays)
primary_images_jpeg # Third-person camera images (JPEG compressed), shape: (T, H, W, 3)
wrist_images_jpeg # Wrist camera images (JPEG compressed), shape: (T, H, W, 3)
proprio # Proprioceptive state (gripper + EEF pos + quat), shape: (T, 9)
actions # Action sequence, shape: (T, 7)
# Attributes (scalars/metadata)
success # Boolean: True if episode succeeded, False otherwise
task_description # String: Natural language task description
Statistics
- Total demonstrations: 500 per task suite before filtering (4 suites = 2000 total)
- Success rate: ~80-90% (varies by task suite)
- Image resolution: 256×256×3 (RGB)
- Action dimensions: 7 (6-DoF end-effector control + 1 gripper)
- Proprioception dimensions: 9 (2 gripper joints + 3 EEF position + 4 EEF quaternion)
Original LIBERO Dataset
This dataset is derived from the original LIBERO benchmark:
- Paper: LIBERO: Benchmarking Knowledge Transfer for Lifelong Robot Learning
- Repository: https://github.com/Lifelong-Robot-Learning/LIBERO
- License: CC BY 4.0
Citation
If you use this dataset, please cite both the original LIBERO paper and the Cosmos Policy paper:
License
Creative Commons Attribution 4.0 International (CC BY 4.0)
- Downloads last month
- 31