robonet

  • Description:

RoboNet contains over 15 million video frames of robot-object interaction, taken from 113 unique camera viewpoints.

  • The actions are deltas in position and rotation to the robot end-effector with one additional dimension of the action vector reserved for the gripper joint.

  • The states are cartesian end-effector control action space with restricted rotation, and a gripper joint

  • Homepage: https://www.robonet.wiki/

  • Source code: tfds.video.Robonet

  • Versions:

    • 4.0.1 (default): No release notes.
  • Supervised keys (See as_supervised doc): None

  • Citation:

@article{dasari2019robonet,
  title={RoboNet: Large-Scale Multi-Robot Learning},
  author={Dasari, Sudeep and Ebert, Frederik and Tian, Stephen and
  Nair, Suraj and Bucher, Bernadette and Schmeckpeper, Karl
  and Singh, Siddharth and Levine, Sergey and Finn, Chelsea},
  journal={arXiv preprint arXiv:1910.11215},
  year={2019}
}

robonet/robonet_sample_64 (default config)

  • Config description: 64x64 RoboNet Sample.

  • Download size: 119.80 MiB

  • Dataset size: 183.04 MiB

  • Auto-cached (documentation): Only when shuffle_files=False (train)

  • Splits:

Split Examples
'train' 700
  • Features:
FeaturesDict({
    'actions': Tensor(shape=(None, 5), dtype=tf.float32),
    'filename': Text(shape=(), dtype=tf.string),
    'states': Tensor(shape=(None, 5), dtype=tf.float32),
    'video': Video(Image(shape=(64, 64, 3), dtype=tf.uint8)),
})

robonet/robonet_sample_128

  • Config description: 128x128 RoboNet Sample.

  • Download size: 119.80 MiB

  • Dataset size: 638.98 MiB

  • Auto-cached (documentation): No

  • Splits:

Split Examples
'train' 700
  • Features:
FeaturesDict({
    'actions': Tensor(shape=(None, 5), dtype=tf.float32),
    'filename': Text(shape=(), dtype=tf.string),
    'states': Tensor(shape=(None, 5), dtype=tf.float32),
    'video': Video(Image(shape=(128, 128, 3), dtype=tf.uint8)),
})

robonet/robonet_64

  • Config description: 64x64 RoboNet.

  • Download size: 36.20 GiB

  • Dataset size: 41.37 GiB

  • Auto-cached (documentation): No

  • Splits:

Split Examples
'train' 162,417
  • Features:
FeaturesDict({
    'actions': Tensor(shape=(None, 5), dtype=tf.float32),
    'filename': Text(shape=(), dtype=tf.string),
    'states': Tensor(shape=(None, 5), dtype=tf.float32),
    'video': Video(Image(shape=(64, 64, 3), dtype=tf.uint8)),
})

robonet/robonet_128

  • Config description: 128x128 RoboNet.

  • Download size: 36.20 GiB

  • Dataset size: 144.90 GiB

  • Auto-cached (documentation): No

  • Splits:

Split Examples
'train' 162,417
  • Features:
FeaturesDict({
    'actions': Tensor(shape=(None, 5), dtype=tf.float32),
    'filename': Text(shape=(), dtype=tf.string),
    'states': Tensor(shape=(None, 5), dtype=tf.float32),
    'video': Video(Image(shape=(128, 128, 3), dtype=tf.uint8)),
})