cmu_stretch

  • תיאור :

שלום אינטראקציות למטבח רובוט מתיחה

לְפַצֵל דוגמאות
'train' 135
  • מבנה תכונה :
FeaturesDict({
    'episode_metadata': FeaturesDict({
        'file_path': Text(shape=(), dtype=string),
    }),
    'steps': Dataset({
        'action': Tensor(shape=(8,), 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({
            'image': Image(shape=(128, 128, 3), dtype=uint8),
            'state': Tensor(shape=(4,), dtype=float32),
        }),
        'reward': Scalar(shape=(), dtype=float32),
    }),
})
  • תיעוד תכונה :
תכונה מעמד צוּרָה Dtype תיאור
FeaturesDict
episode_metadata FeaturesDict
episode_metadata/file_path טֶקסט חוּט נתיב לקובץ הנתונים המקורי.
צעדים מערך נתונים
צעדים/פעולה מוֹתֵחַ (8,) לצוף32 פעולת רובוט, מורכבת מ[3x ee pos, 3x ee rot 1x gripper action binary, 1x terminate episode].
צעדים/הנחה סקלר לצוף32 הנחה אם ניתנת, ברירת המחדל היא 1.
צעדים/הוא_ראשון מוֹתֵחַ bool
צעדים/הוא_אחרון מוֹתֵחַ bool
steps/is_terminal מוֹתֵחַ bool
שלבים/הטבעת_שפה מוֹתֵחַ (512,) לצוף32 הטבעת שפת Kona. ראה https://tfhub.dev/google/universal-sentence-encoder-large/5
שלבים/הוראת_שפה טֶקסט חוּט הוראת שפה.
צעדים/תצפית FeaturesDict
צעדים/תצפית/תמונה תמונה (128, 128, 3) uint8 תצפית RGB של מצלמה ראשית.
צעדים/תצפית/מצב מוֹתֵחַ (4,) לצוף32 מצב רובוט, מורכב מ[3x זוויות מפרק רובוט/ee pos, 1x מיקום תפס].
צעדים/פרס סקלר לצוף32 תגמול אם מסופק, 1 בשלב האחרון להדגמות.
  • ציטוט :
@inproceedings{bahl2023affordances,
  title={Affordances from Human Videos as a Versatile Representation for Robotics},
  author={Bahl, Shikhar and Mendonca, Russell and Chen, Lili and Jain, Unnat and Pathak, Deepak},
  booktitle={CVPR},
  year={2023}
}
@article{mendonca2023structured,
  title={Structured World Models from Human Videos},
  author={Mendonca, Russell and Bahl, Shikhar and Pathak, Deepak},
  journal={CoRL},
  year={2023}
}