utokyo_xarm_bimanual_converted_externally_to_rlds

  • 説明

xArm 両手セットアップ折りたたみタオル

スプリット
'train' 64
'val' 6
  • 機能の構造:
FeaturesDict({
    'episode_metadata': FeaturesDict({
        'file_path': Text(shape=(), dtype=string),
    }),
    'steps': Dataset({
        'action': Tensor(shape=(14,), 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': Text(shape=(), dtype=string),
        'observation': FeaturesDict({
            'action_l': Tensor(shape=(7,), dtype=float32),
            'action_r': Tensor(shape=(7,), dtype=float32),
            'image': Image(shape=(256, 256, 3), dtype=uint8),
            'pose_l': Tensor(shape=(6,), dtype=float32),
            'pose_r': Tensor(shape=(6,), 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/ファイルパス文章元のデータ ファイルへのパス。
ステップデータセット
ステップ/アクションテンソル(14,) float32ロボット アクションは、[3x EEF 位置 (L)、3x EEF 方向のヨー/ピッチ/ロール (L)、1x グリッパー開閉位置 (L)、3x EEF 位置 (R)、3x EEF 方向のヨー/ピッチ/ロールで構成されます。 (R)、1x グリッパー開閉位置 (R)]。
歩数/割引スカラーfloat32割引が指定されている場合、デフォルトは 1 です。
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コナ言語の埋め込み。 https://tfhub.dev/google/universal-sentence-encoder-large/5を参照してください。
ステップ/言語説明文章言語指導。
ステップ/観察特徴辞書
ステップ/観察/アクション_lテンソル(7、) float32左側のロボット アクションは、[3x EEF 位置、3x EEF 方向のヨー/ピッチ/ロール] で構成されます。
ステップ/観察/アクション_rテンソル(7、) float32右のロボット アクションは、[3x EEF 位置、3x EEF 方向ヨー/ピッチ/ロール、1x グリッパー開閉位置] で構成されます。
手順・観察・イメージ画像(256, 256, 3) uint8メインカメラRGB観察。
歩数/観察/pose_lテンソル(6,) float32左側のロボット エンド エフェクターのポーズは、[3x EEF 位置、3x EEF 方向のヨー/ピッチ/ロール] で構成されます。
歩数/観察/pose_rテンソル(6,) float32右のロボット エンド エフェクタのポーズは、[3x EEF 位置、3x EEF 方向のヨー/ピッチ/ロール] で構成されます。
歩数/報酬スカラーfloat32提供されている場合は報酬、デモの最終ステップで 1。
@misc{matsushima2023weblab,
  title={Weblab xArm Dataset},
  author={Tatsuya Matsushima and Hiroki Furuta and Yusuke Iwasawa and Yutaka Matsuo},
  year={2023},
}