- 説明:
MT-Opt 論文のデータセット。
ホームページ: https ://karolhausman.github.io/mt-opt/
ソース コード:
tfds.robotics.mt_opt.MtOptバージョン:
-
1.0.0(デフォルト): 初期リリース。
-
ダウンロードサイズ: サイズ
Unknown size自動キャッシュ(ドキュメント): いいえ
監視されたキー(
as_superviseddocを参照):None図( tfds.show_examples ): サポートされていません。
引用:
@misc{kalashnikov2021mtopt,
title={MT-Opt: Continuous Multi-Task Robotic Reinforcement Learning at Scale},
author={Dmitry Kalashnikov and Jacob Varley and Yevgen Chebotar and Benjamin Swanson and Rico Jonschkowski and Chelsea Finn and Sergey Levine and Karol Hausman},
year={2021},
eprint={2104.08212},
archivePrefix={arXiv},
primaryClass={cs.RO}
}
mt_opt/rlds (デフォルト設定)
構成の説明: このデータセットには、実際のロボットの艦隊全体で収集されたタスク エピソードが含まれています。ステップとエピソードを表すためにRLDS 形式に従います。
データセットのサイズ:
4.38 TiBスプリット:
| スプリット | 例 |
|---|---|
'train' | 920,165 |
- 機能構造:
FeaturesDict({
'episode_id': string,
'skill': uint8,
'steps': Dataset({
'action': FeaturesDict({
'close_gripper': bool,
'open_gripper': bool,
'target_pose': Tensor(shape=(7,), dtype=float32),
'terminate': bool,
}),
'is_first': bool,
'is_last': bool,
'is_terminal': bool,
'observation': FeaturesDict({
'gripper_closed': bool,
'height_to_bottom': float32,
'image': Image(shape=(512, 640, 3), dtype=uint8),
'state_dense': Tensor(shape=(7,), dtype=float32),
}),
}),
'task_code': string,
})
- 機能のドキュメント:
| 特徴 | クラス | 形 | Dtype | 説明 |
|---|---|---|---|---|
| 特徴辞書 | ||||
| episode_id | テンソル | ストリング | ||
| スキル | テンソル | uint8 | ||
| ステップ | データセット | |||
| ステップ/アクション | 特徴辞書 | |||
| ステップ/アクション/close_gripper | テンソル | ブール | ||
| ステップ/アクション/open_gripper | テンソル | ブール | ||
| ステップ/アクション/ターゲット_ポーズ | テンソル | (7、) | float32 | |
| ステップ/アクション/終了 | テンソル | ブール | ||
| ステップ/is_first | テンソル | ブール | ||
| ステップ/is_last | テンソル | ブール | ||
| ステップ/is_terminal | テンソル | ブール | ||
| ステップ/観察 | 特徴辞書 | |||
| ステップ/観察/gripper_closed | テンソル | ブール | ||
| steps/observation/height_to_bottom | テンソル | float32 | ||
| 歩み・観察・イメージ | 画像 | (512、640、3) | uint8 | |
| ステップ/観察/state_dense | テンソル | (7、) | float32 | |
| タスクコード | テンソル | ストリング |
- 例( tfds.as_dataframe ):
mt_opt/sd
構成の説明: 人がキュレートしたタスク完了の定義を含む成功検出器データセット。
データセットサイズ:
548.56 GiBスプリット:
| スプリット | 例 |
|---|---|
'test' | 94,636 |
'train' | 380,234 |
- 機能構造:
FeaturesDict({
'image_0': Image(shape=(512, 640, 3), dtype=uint8),
'image_1': Image(shape=(480, 640, 3), dtype=uint8),
'image_2': Image(shape=(480, 640, 3), dtype=uint8),
'success': bool,
'task_code': string,
})
- 機能のドキュメント:
| 特徴 | クラス | 形 | Dtype | 説明 |
|---|---|---|---|---|
| 特徴辞書 | ||||
| 画像_0 | 画像 | (512、640、3) | uint8 | |
| 画像_1 | 画像 | (480、640、3) | uint8 | |
| 画像_2 | 画像 | (480、640、3) | uint8 | |
| 成功 | テンソル | ブール | ||
| タスクコード | テンソル | ストリング |
- 例( tfds.as_dataframe ):