utokyo_xarm_pick_and_place_converted_externally_to_rlds

  • 설명 :

xArm 개체 선택 및 배치

나뉘다
'train' 92
'val' 10
  • 기능 구조 :
FeaturesDict({
    'episode_metadata': FeaturesDict({
        'file_path': Text(shape=(), dtype=string),
    }),
    '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': Text(shape=(), dtype=string),
        'observation': FeaturesDict({
            'end_effector_pose': Tensor(shape=(6,), dtype=float32),
            'hand_image': Image(shape=(224, 224, 3), dtype=uint8),
            'image': Image(shape=(224, 224, 3), dtype=uint8),
            'image2': Image(shape=(224, 224, 3), dtype=uint8),
            'joint_state': Tensor(shape=(14,), dtype=float32),
            'joint_trajectory': Tensor(shape=(21,), dtype=float32),
        }),
        'reward': Scalar(shape=(), dtype=float32),
    }),
})
  • 기능 문서 :
특징 수업 모양 Dtype 설명
특징Dict
에피소드_메타데이터 특징Dict
에피소드_메타데이터/파일_경로 텍스트 원본 데이터 파일의 경로입니다.
단계 데이터세트
단계/작업 텐서 (7,) float32 로봇 동작은 [3x EEF 위치, 3x EEF 방향 요/피치/롤, 1x 그리퍼 열기/닫기 위치]로 구성됩니다.
걸음수/할인 스칼라 float32 할인이 제공되면 기본값은 1입니다.
걸음수/is_first 텐서 부울
걸음수/is_last 텐서 부울
단계/is_terminal 텐서 부울
단계/언어_임베딩 텐서 (512,) float32 코나 언어 임베딩. https://tfhub.dev/google/universal-sentence-encoder-large/5를 참조하세요.
단계/언어_지시 텍스트 언어 교육.
단계/관찰 특징Dict
단계/관찰/end_effor_pose 텐서 (6,) float32 로봇 엔드 이펙터 포즈는 [3x EEF 위치, 3x EEF 방향 요/피치/롤]로 구성됩니다.
단계/관찰/손_이미지 영상 (224, 224, 3) uint8 핸드카메라 RGB 관찰.
단계/관찰/이미지 영상 (224, 224, 3) uint8 메인 카메라 RGB 관찰.
단계/관찰/이미지2 영상 (224, 224, 3) uint8 다른 시점에서 또 다른 카메라 RGB 관찰.
단계/관찰/joint_state 텐서 (14,) float32 로봇 관절 상태는 [7x 로봇 관절 각도, 7x 로봇 관절 속도]로 구성됩니다.
단계/관찰/joint_trajectory 텐서 (21,) float32 로봇 관절 궤적은 [7x 로봇 관절 각도, 7x 로봇 관절 속도, 7x 로봇 관절 가속도]로 구성됩니다.
걸음수/보상 스칼라 float32 제공되는 경우 보상, 데모의 마지막 단계에서 1개.
@misc{matsushima2023weblab,
  title={Weblab xArm Dataset},
  author={Tatsuya Matsushima and Hiroki Furuta and Yusuke Iwasawa and Yutaka Matsuo},
  year={2023},
}