ブリッジデータ_msr

  • 説明

UC Berkeley の BridgeData V2 ( https://rail-berkeley.github.io/bridgedata/ ) と互換性のあるセットアップと形式で、WidowX-250 ロボットに関して Microsoft Research で収集された一連のオブジェクト操作軌跡

スプリット
'train' 822
  • 機能の構造:
FeaturesDict({
    'episode_metadata': FeaturesDict({
        'file_path': string,
        'has_depth_0': Scalar(shape=(), dtype=bool),
        'has_image_0': Scalar(shape=(), dtype=bool),
        'has_image_1': Scalar(shape=(), dtype=bool),
        'has_image_2': Scalar(shape=(), dtype=bool),
        'has_language': Scalar(shape=(), dtype=bool),
    }),
    'steps': Dataset({
        'action': Tensor(shape=(7,), dtype=float32),
        'discount': Scalar(shape=(), dtype=float32),
        'is_first': bool,
        'is_last': bool,
        'is_terminal': bool,
        'language_embedding': Tensor(shape=(512,), dtype=float32),
        'language_instruction': string,
        'observation': FeaturesDict({
            'depth_0': Image(shape=(480, 640, 1), dtype=uint8),
            'image_0': Image(shape=(480, 640, 3), dtype=uint8),
            'image_1': Image(shape=(480, 640, 3), dtype=uint8),
            'image_2': Image(shape=(480, 640, 3), dtype=uint8),
            'state': Tensor(shape=(7,), dtype=float32),
        }),
        'reward': Scalar(shape=(), dtype=float32),
    }),
})
  • 機能ドキュメント:
特徴クラスDタイプ説明
特徴辞書
エピソード_メタデータ特徴辞書
エピソードメタデータ/ファイルパステンソル
エピソードメタデータ/has_ Depth_0スカラーブール
エピソードメタデータ/has_image_0スカラーブール
エピソード_メタデータ/has_image_1スカラーブール
エピソードメタデータ/has_image_2スカラーブール
エピソードのメタデータ/言語ありスカラーブール
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/割引スカラーfloat32
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/言語_埋め込みテンソル(512,) float32
ステップ/言語説明テンソル
ステップ/観察特徴辞書
歩数/観察/深さ_0画像(480、640、1) uint8
歩数/観察/image_0画像(480、640、3) uint8
ステップ/観察/画像_1画像(480、640、3) uint8
ステップ/観察/画像_2画像(480、640、3) uint8
ステップ/観察/状態テンソル(7、) float32
歩数/報酬スカラーfloat32
  • 引用