austin_sailor_dataset_converted_externally_to_rlds

  • Descrição :

Tarefas de mesa Franka

Dividir Exemplos
'train' 240
  • Estrutura de recursos :
FeaturesDict({
   
'episode_metadata': FeaturesDict({
       
'file_path': Text(shape=(), dtype=string),
   
}),
   
'steps': Dataset({
       
'action': Tensor(shape=(7,), dtype=float32, description=Robot action, consists of [3x ee relative pos, 3x ee relative rotation, 1x gripper action].),
       
'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({
           
'image': Image(shape=(128, 128, 3), dtype=uint8, description=Main camera RGB observation.),
           
'state': Tensor(shape=(8,), dtype=float32, description=Default robot state, consists of [3x robot ee pos, 3x ee quat, 1x gripper state].),
           
'state_ee': Tensor(shape=(16,), dtype=float32, description=End-effector state, represented as 4x4 homogeneous transformation matrix of ee pose.),
           
'state_gripper': Tensor(shape=(1,), dtype=float32, description=Robot gripper opening width. Ranges between ~0 (closed) to ~0.077 (open)),
           
'state_joint': Tensor(shape=(7,), dtype=float32, description=Robot 7-dof joint information (not used in original SAILOR dataset).),
           
'wrist_image': Image(shape=(128, 128, 3), dtype=uint8, description=Wrist camera RGB observation.),
       
}),
       
'reward': Scalar(shape=(), dtype=float32, description=True on last step of the episode.),
   
}),
})
  • Documentação de recursos :
Recurso Aula Forma Tipo D Descrição
RecursosDict
episódio_metadados RecursosDict
episódio_metadados/caminho_do_arquivo Texto corda Caminho para o arquivo de dados original.
passos Conjunto de dados
etapas/ação Tensor (7,) float32 A ação do robô consiste em [3x ee posição relativa, 3x ee rotação relativa, 1x ação da pinça].
passos/desconto Escalar float32 Desconto, se fornecido, o padrão é 1.
passos/é_primeiro Tensor bool
passos/é_último Tensor bool
etapas/is_terminal Tensor bool
etapas/idioma_incorporação Tensor (512,) float32 Incorporação da linguagem Kona. Consulte https://tfhub.dev/google/universal-sentence-encoder-large/5
etapas/instrução_idioma Texto corda Instrução de Idiomas.
etapas/observação RecursosDict
passos/observação/imagem Imagem (128, 128, 3) uint8 Observação RGB da câmera principal.
etapas/observação/estado Tensor (8,) float32 O estado padrão do robô consiste em [3x robô ee pos, 3x ee quat, 1x estado da garra].
etapas/observação/estado_ee Tensor (16,) float32 Estado efetor final, representado como matriz de transformação homogênea 4x4 de ee pose.
etapas/observação/state_gripper Tensor (1,) float32 Largura de abertura da garra do robô. Varia entre ~0 (fechado) a ~0,077 (aberto)
etapas/observação/state_joint Tensor (7,) float32 Informações conjuntas do robô 7-dof (não usadas no conjunto de dados SAILOR original).
passos/observação/imagem_de_pulso Imagem (128, 128, 3) uint8 Observação RGB da câmera de pulso.
passos/recompensa Escalar float32 Verdade na última etapa do episódio.
  • Citação :
@inproceedings{nasiriany2022sailor,
      title
={Learning and Retrieval from Prior Data for Skill-based Imitation Learning},
      author
={Soroush Nasiriany and Tian Gao and Ajay Mandlekar and Yuke Zhu},
      booktitle
={Conference on Robot Learning (CoRL)},
      year
={2022}
   
}