utokyo_saytap_converted_externally_to_rlds

  • 설명 :

A1 워킹, RGB 없음

나뉘다
'train' 20
  • 기능 구조 :
FeaturesDict({
    'episode_metadata': FeaturesDict({
        'file_path': Text(shape=(), dtype=string),
    }),
    'steps': Dataset({
        'action': Tensor(shape=(12,), 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({
            'desired_pattern': Tensor(shape=(4, 5), dtype=bool),
            'desired_vel': Tensor(shape=(3,), dtype=float32),
            'image': Image(shape=(64, 64, 3), dtype=uint8),
            'prev_act': Tensor(shape=(12,), dtype=float32),
            'proj_grav_vec': Tensor(shape=(3,), dtype=float32),
            'state': Tensor(shape=(30,), dtype=float32),
            'wrist_image': Image(shape=(64, 64, 3), dtype=uint8),
        }),
        '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 (12,) float32 로봇 액션은 [12x 관절 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/원하는_패턴 텐서 (4, 5) 부울 4개 다리에 대해 원하는 발 접촉 패턴, 4개 행은 앞 오른쪽, 앞 왼쪽, 뒤 오른쪽 및 뒤 왼쪽 다리에 대한 것이며 패턴 길이는 5(=0.1s)입니다.
단계/관찰/desired_vel 텐서 (삼,) float32 원하는 속도. 처음 2개는 진행 방향에 수직인 선형 속도이고, 세 번째는 요 축에 대한 원하는 각속도입니다.
단계/관찰/이미지 영상 (64, 64, 3) uint8 더미 카메라 RGB 관찰.
단계/관찰/prev_act 텐서 (12,) float32 이전 단계에서 적용된 작업입니다.
단계/관찰/proj_grav_vec 텐서 (삼,) float32 로봇 기본 프레임의 중력 벡터 [0, 0, -1]입니다.
단계/관찰/상태 텐서 (30,) float32 로봇 상태는 [3x 로봇 기본 선형 속도, 3x 기본 각도 벨, 12x 관절 위치, 12x 관절 속도]로 구성됩니다.
걸음수/관찰/wrist_image 영상 (64, 64, 3) uint8 더미 손목 카메라 RGB 관찰.
걸음수/보상 스칼라 float32 제공되는 경우 보상, 데모의 마지막 단계에서 1개.
@article{saytap2023,
  author = {Yujin Tang and Wenhao Yu and Jie Tan and Heiga Zen and Aleksandra Faust and
Tatsuya Harada},
  title  = {SayTap: Language to Quadrupedal Locomotion},
  eprint = {arXiv:2306.07580},
  url    = {https://saytap.github.io},
  note   = "{https://saytap.github.io}",
  year   = {2023}
}