stanford_mask_vit_converted_externally_to_rlds

  • 説明

ゴミ箱の中の物体を押したり選んだりするソーヤー

スプリット
'train' 9,109
'val' 91
  • 機能の構造:
FeaturesDict({
    'episode_metadata': FeaturesDict({
        'file_path': Text(shape=(), dtype=string),
    }),
    'steps': Dataset({
        'action': Tensor(shape=(5,), dtype=float32, description=Robot action, consists of [3x change in end effector position, 1x gripper yaw, 1x open/close gripper (-1 means to open the gripper, 1 means close)].),
        'discount': Scalar(shape=(), dtype=float32, description=Discount if provided, default to 1.),
        'is_first': bool,
        'is_last': bool,
        'is_terminal': bool,
        'language_embedding': Tensor(shape=(512,), dtype=float32, description=Kona language embedding. See https://tfhub.dev/google/universal-sentence-encoder-large/5),
        'language_instruction': Text(shape=(), dtype=string),
        'observation': FeaturesDict({
            'end_effector_pose': Tensor(shape=(5,), dtype=float32, description=Robot end effector pose, consists of [3x Cartesian position, 1x gripper yaw, 1x gripper position]. This is the state used in the MaskViT paper.),
            'finger_sensors': Tensor(shape=(1,), dtype=float32, description=1x Sawyer gripper finger sensors.),
            'high_bound': Tensor(shape=(5,), dtype=float32, description=High bound for end effector pose normalization. Consists of [3x Cartesian position, 1x gripper yaw, 1x gripper position].),
            'image': Image(shape=(480, 480, 3), dtype=uint8, description=Main camera RGB observation.),
            'low_bound': Tensor(shape=(5,), dtype=float32, description=Low bound for end effector pose normalization. Consists of [3x Cartesian position, 1x gripper yaw, 1x gripper position].),
            'state': Tensor(shape=(15,), dtype=float32, description=Robot state, consists of [7x robot joint angles, 7x robot joint velocities,1x gripper position].),
        }),
        'reward': Scalar(shape=(), dtype=float32, description=Reward if provided, 1 on final step for demos.),
    }),
})
  • 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(5,) float32ロボットのアクションは、[エンド エフェクターの位置の変更 3 回、グリッパーのヨー 1 回、グリッパーの開閉 1 回 (-1 はグリッパーを開くことを意味し、1 は閉じることを意味します)] で構成されます。
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/end_effect_poseテンソル(5,) float32ロボット エンド エフェクターのポーズは、[3x デカルト位置、1x グリッパー ヨー、1x グリッパー位置] で構成されます。これはMaskViT論文で使用されている状態です。
歩数/観察/指センサーテンソル(1,) float32 1x Sawyer グリッパーフィンガーセンサー。
ステップ/観察/高境界テンソル(5,) float32エンドエフェクターのポーズ正規化の上限。 [3x デカルト位置、1x グリッパー ヨー、1x グリッパー位置] で構成されます。
手順・観察・イメージ画像(480、480、3) uint8メインカメラRGB観察。
ステップ/観察/low_boundテンソル(5,) float32エンドエフェクターのポーズ正規化の下限。 [3x デカルト位置、1x グリッパー ヨー、1x グリッパー位置] で構成されます。
ステップ/観察/状態テンソル(15,) float32ロボットの状態は、[7x ロボット関節角度、7x ロボット関節速度、1x グリッパー位置] 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。
@inproceedings{gupta2022maskvit,
  title={MaskViT: Masked Visual Pre-Training for Video Prediction},
  author={Agrim Gupta and Stephen Tian and Yunzhi Zhang and Jiajun Wu and Roberto Martín-Martín and Li Fei-Fei},
  booktitle={International Conference on Learning Representations},
  year={2022}
}