robosuite_panda_pick_place_can

  • 説明

これらのデータセットは、robosuite ロボット アーム シミュレーターの PickPlaceCan 環境を使用して作成されました。人間のデータセットは、 RLDS Creatorとゲームパッド コントローラーを使用して 1 人のオペレーターによって記録されました。

合成データセットは、 EnvLogger ライブラリを使用して記録されています。

データセットはRLDS 形式に従ってステップとエピソードを表します。

エピソードは 400 のステップで構成されます。各エピソードでは、タスクが完了するとタグが追加され、このタグはカスタム ステップ メタデータの一部として保存されます。

EnvLogger の依存関係により、このデータセットの生成は現在 Linux 環境でのみサポートされていることに注意してください。

@misc{ramos2021rlds,
      title={RLDS: an Ecosystem to Generate, Share and Use Datasets in Reinforcement Learning},
      author={Sabela Ramos and Sertan Girgin and Léonard Hussenot and Damien Vincent and Hanna Yakubovich and Daniel Toyama and Anita Gergely and Piotr Stanczyk and Raphael Marinier and Jeremiah Harmsen and Olivier Pietquin and Nikola Momchev},
      year={2021},
      eprint={2111.02767},
      archivePrefix={arXiv},
      primaryClass={cs.LG}
}

robosuite_panda_pick_place_can/human_dc29b40a (デフォルト設定)

  • 構成の説明: 人間が生成したデータセット (50 エピソード)。

  • ホームページ: https://github.com/google-research/rlds

  • ダウンロードサイズ: 96.67 MiB

  • データセットのサイズ: 407.24 MiB

  • 分割:

スプリット
'train' 50
  • 機能の構造:
FeaturesDict({
    'agent_id': string,
    'episode_id': string,
    'episode_index': int32,
    'steps': Dataset({
        'action': Tensor(shape=(7,), dtype=float64),
        'discount': float64,
        'image': Image(shape=(None, None, 3), dtype=uint8),
        'is_first': bool,
        'is_last': bool,
        'is_terminal': bool,
        'observation': FeaturesDict({
            'Can_pos': Tensor(shape=(3,), dtype=float64),
            'Can_quat': Tensor(shape=(4,), dtype=float64),
            'Can_to_robot0_eef_pos': Tensor(shape=(3,), dtype=float64),
            'Can_to_robot0_eef_quat': Tensor(shape=(4,), dtype=float32),
            'object-state': Tensor(shape=(14,), dtype=float64),
            'robot0_eef_pos': Tensor(shape=(3,), dtype=float64),
            'robot0_eef_quat': Tensor(shape=(4,), dtype=float64),
            'robot0_gripper_qpos': Tensor(shape=(2,), dtype=float64),
            'robot0_gripper_qvel': Tensor(shape=(2,), dtype=float64),
            'robot0_joint_pos_cos': Tensor(shape=(7,), dtype=float64),
            'robot0_joint_pos_sin': Tensor(shape=(7,), dtype=float64),
            'robot0_joint_vel': Tensor(shape=(7,), dtype=float64),
            'robot0_proprio-state': Tensor(shape=(32,), dtype=float64),
        }),
        'reward': float64,
        'tag:placed': bool,
    }),
})
  • 機能ドキュメント:
特徴クラスDタイプ説明
特徴辞書
エージェントIDテンソル
エピソードIDテンソル
エピソードインデックステンソルint32
ステップデータセット
ステップ/アクションテンソル(7、) float64
歩数/割引テンソルfloat64
ステップ/イメージ画像(なし、なし、3) uint8
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
歩数/観察/Can_posテンソル(3,) float64
ステップ/観察/Can_quatテンソル(4,) float64
ステップ/観察/Can_to_robot0_eef_posテンソル(3,) float64
ステップ/観察/Can_to_robot0_eef_quatテンソル(4,) float32
ステップ/観察/オブジェクトの状態テンソル(14,) float64
ステップ/観察/robot0_eef_posテンソル(3,) float64
ステップ/観察/robot0_eef_quatテンソル(4,) float64
ステップ/観察/robot0_gripper_qposテンソル(2,) float64
ステップ/観察/robot0_gripper_qvelテンソル(2,) float64
ステップ/観察/robot0_joint_pos_cosテンソル(7、) float64
ステップ/観察/robot0_joint_pos_sinテンソル(7、) float64
ステップ/観察/robot0_joint_velテンソル(7、) float64
ステップ/観察/robot0_proprio-stateテンソル(32,) float64
歩数/報酬テンソルfloat64
ステップ/タグ:配置されましたテンソルブール

robosuite_panda_pick_place_can/human_images_dc29b40a

  • 構成の説明: 人間が生成したデータセット。観察時のさまざまなカメラ角度の画像が含まれます。生成には時間がかかる場合がありますのでご了承ください。

  • ホームページ: https://www.tensorflow.org/datasets/catalog/robosuite_panda_pick_place_can

  • ダウンロードサイズ: 10.95 GiB

  • データセットのサイズ: 7.53 GiB

  • 分割:

スプリット
'train' 50
  • 機能の構造:
FeaturesDict({
    'agent_id': string,
    'episode_id': string,
    'episode_index': int32,
    'steps': Dataset({
        'action': Tensor(shape=(7,), dtype=float64),
        'discount': float64,
        'image': Image(shape=(None, None, 3), dtype=uint8),
        'is_first': bool,
        'is_last': bool,
        'is_terminal': bool,
        'observation': FeaturesDict({
            'Can_pos': Tensor(shape=(3,), dtype=float64),
            'Can_quat': Tensor(shape=(4,), dtype=float64),
            'Can_to_robot0_eef_pos': Tensor(shape=(3,), dtype=float64),
            'Can_to_robot0_eef_quat': Tensor(shape=(4,), dtype=float32),
            'agentview_image': Image(shape=(256, 256, 3), dtype=uint8),
            'birdview_image': Image(shape=(256, 256, 3), dtype=uint8),
            'object-state': Tensor(shape=(14,), dtype=float64),
            'robot0_eef_pos': Tensor(shape=(3,), dtype=float64),
            'robot0_eef_quat': Tensor(shape=(4,), dtype=float64),
            'robot0_eye_in_hand_image': Image(shape=(256, 256, 3), dtype=uint8),
            'robot0_gripper_qpos': Tensor(shape=(2,), dtype=float64),
            'robot0_gripper_qvel': Tensor(shape=(2,), dtype=float64),
            'robot0_joint_pos_cos': Tensor(shape=(7,), dtype=float64),
            'robot0_joint_pos_sin': Tensor(shape=(7,), dtype=float64),
            'robot0_joint_vel': Tensor(shape=(7,), dtype=float64),
            'robot0_proprio-state': Tensor(shape=(32,), dtype=float64),
            'robot0_robotview_image': Image(shape=(256, 256, 3), dtype=uint8),
        }),
        'reward': float64,
        'tag:placed': bool,
    }),
})
  • 機能ドキュメント:
特徴クラスDタイプ説明
特徴辞書
エージェントIDテンソル
エピソードIDテンソル
エピソードインデックステンソルint32
ステップデータセット
ステップ/アクションテンソル(7、) float64
歩数/割引テンソルfloat64
ステップ/イメージ画像(なし、なし、3) uint8
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
歩数/観察/Can_posテンソル(3,) float64
ステップ/観察/Can_quatテンソル(4,) float64
ステップ/観察/Can_to_robot0_eef_posテンソル(3,) float64
ステップ/観察/Can_to_robot0_eef_quatテンソル(4,) float32
ステップ/観察/agentview_image画像(256, 256, 3) uint8
歩数/観察/birdview_image画像(256, 256, 3) uint8
ステップ/観察/オブジェクトの状態テンソル(14,) float64
ステップ/観察/robot0_eef_posテンソル(3,) float64
ステップ/観察/robot0_eef_quatテンソル(4,) float64
ステップ/観察/robot0_eye_in_hand_image画像(256, 256, 3) uint8
ステップ/観察/robot0_gripper_qposテンソル(2,) float64
ステップ/観察/robot0_gripper_qvelテンソル(2,) float64
ステップ/観察/robot0_joint_pos_cosテンソル(7、) float64
ステップ/観察/robot0_joint_pos_sinテンソル(7、) float64
ステップ/観察/robot0_joint_velテンソル(7、) float64
ステップ/観察/robot0_proprio-stateテンソル(32,) float64
ステップ/観察/robot0_robotview_image画像(256, 256, 3) uint8
歩数/報酬テンソルfloat64
ステップ/タグ:配置されましたテンソルブール

robosuite_panda_pick_place_can/synthetic_stochastic_sac_afe13968

  • 構成の説明: SAC でトレーニングされた確率的エージェントによって生成された合成データセット (200 エピソード)。

  • ホームページ: https://github.com/google-research/rlds

  • ダウンロードサイズ: 144.44 MiB

  • データセットのサイズ: 622.86 MiB

  • 分割:

スプリット
'train' 200
  • 機能の構造:
FeaturesDict({
    'agent_id': string,
    'episode_id': string,
    'episode_index': int32,
    'steps': Dataset({
        'action': Tensor(shape=(7,), dtype=float32),
        'discount': float64,
        'image': Image(shape=(None, None, 3), dtype=uint8),
        'is_first': bool,
        'is_last': bool,
        'is_terminal': bool,
        'observation': FeaturesDict({
            'Can_pos': Tensor(shape=(3,), dtype=float32),
            'Can_quat': Tensor(shape=(4,), dtype=float32),
            'Can_to_robot0_eef_pos': Tensor(shape=(3,), dtype=float32),
            'Can_to_robot0_eef_quat': Tensor(shape=(4,), dtype=float32),
            'object-state': Tensor(shape=(14,), dtype=float32),
            'robot0_eef_pos': Tensor(shape=(3,), dtype=float32),
            'robot0_eef_quat': Tensor(shape=(4,), dtype=float32),
            'robot0_gripper_qpos': Tensor(shape=(2,), dtype=float32),
            'robot0_gripper_qvel': Tensor(shape=(2,), dtype=float32),
            'robot0_joint_pos_cos': Tensor(shape=(7,), dtype=float32),
            'robot0_joint_pos_sin': Tensor(shape=(7,), dtype=float32),
            'robot0_joint_vel': Tensor(shape=(7,), dtype=float32),
            'robot0_proprio-state': Tensor(shape=(32,), dtype=float32),
        }),
        'reward': float64,
        'tag:placed': bool,
    }),
})
  • 機能ドキュメント:
特徴クラスDタイプ説明
特徴辞書
エージェントIDテンソル
エピソードIDテンソル
エピソードインデックステンソルint32
ステップデータセット
ステップ/アクションテンソル(7、) float32
歩数/割引テンソルfloat64
ステップ/イメージ画像(なし、なし、3) uint8
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
歩数/観察/Can_posテンソル(3,) float32
ステップ/観察/Can_quatテンソル(4,) float32
ステップ/観察/Can_to_robot0_eef_posテンソル(3,) float32
ステップ/観察/Can_to_robot0_eef_quatテンソル(4,) float32
ステップ/観察/オブジェクトの状態テンソル(14,) float32
ステップ/観察/robot0_eef_posテンソル(3,) float32
ステップ/観察/robot0_eef_quatテンソル(4,) float32
ステップ/観察/robot0_gripper_qposテンソル(2,) float32
ステップ/観察/robot0_gripper_qvelテンソル(2,) float32
ステップ/観察/robot0_joint_pos_cosテンソル(7、) float32
ステップ/観察/robot0_joint_pos_sinテンソル(7、) float32
ステップ/観察/robot0_joint_velテンソル(7、) float32
ステップ/観察/robot0_proprio-stateテンソル(32,) float32
歩数/報酬テンソルfloat64
ステップ/タグ:配置されましたテンソルブール