taco_play

  • Description:

Franka arm interacting with kitchen

Split Examples
'test' 361
'train' 3,242
  • Feature structure:
FeaturesDict({
    'steps': Dataset({
        'action': FeaturesDict({
            'actions': Tensor(shape=(7,), dtype=float32, description=absolute desired values for gripper pose (first 6 dimensions are x, y, z, yaw, pitch, roll), last dimension is open_gripper (-1 is open gripper, 1 is close)),
            'rel_actions_gripper': Tensor(shape=(7,), dtype=float32, description=relative actions for gripper pose in the gripper camera frame (first 6 dimensions are x, y, z, yaw, pitch, roll), last dimension is open_gripper (-1 is open gripper, 1 is close)),
            'rel_actions_world': Tensor(shape=(7,), dtype=float32, description=relative actions for gripper pose in the robot base frame (first 6 dimensions are x, y, z, yaw, pitch, roll), last dimension is open_gripper (-1 is open gripper, 1 is close)),
            'terminate_episode': float32,
        }),
        'is_first': bool,
        'is_last': bool,
        'is_terminal': bool,
        'observation': FeaturesDict({
            'depth_gripper': Tensor(shape=(84, 84), dtype=float32),
            'depth_static': Tensor(shape=(150, 200), dtype=float32),
            'natural_language_embedding': Tensor(shape=(512,), dtype=float32),
            'natural_language_instruction': string,
            'rgb_gripper': Image(shape=(84, 84, 3), dtype=uint8),
            'rgb_static': Image(shape=(150, 200, 3), dtype=uint8, description=RGB static image of shape. (150, 200, 3). Subsampled from (200,200, 3) image.),
            'robot_obs': Tensor(shape=(15,), dtype=float32, description=EE position (3), EE orientation in euler angles (3), gripper width (1), joint positions (7), gripper action (1)),
            'structured_language_instruction': string,
        }),
        'reward': Scalar(shape=(), dtype=float32),
    }),
})
  • Feature documentation:
Feature Class Shape Dtype Description
FeaturesDict
steps Dataset
steps/action FeaturesDict
steps/action/actions Tensor (7,) float32 absolute desired values for gripper pose (first 6 dimensions are x, y, z, yaw, pitch, roll), last dimension is open_gripper (-1 is open gripper, 1 is close)
steps/action/rel_actions_gripper Tensor (7,) float32 relative actions for gripper pose in the gripper camera frame (first 6 dimensions are x, y, z, yaw, pitch, roll), last dimension is open_gripper (-1 is open gripper, 1 is close)
steps/action/rel_actions_world Tensor (7,) float32 relative actions for gripper pose in the robot base frame (first 6 dimensions are x, y, z, yaw, pitch, roll), last dimension is open_gripper (-1 is open gripper, 1 is close)
steps/action/terminate_episode Tensor float32
steps/is_first Tensor bool
steps/is_last Tensor bool
steps/is_terminal Tensor bool
steps/observation FeaturesDict
steps/observation/depth_gripper Tensor (84, 84) float32
steps/observation/depth_static Tensor (150, 200) float32
steps/observation/natural_language_embedding Tensor (512,) float32
steps/observation/natural_language_instruction Tensor string Natural language instruction is a natural language instruction randomly sampled based on potential task synonyms derived from the structured language task. For example, 'turn blue light off' may map to 'switch the blue color light to off'.
steps/observation/rgb_gripper Image (84, 84, 3) uint8
steps/observation/rgb_static Image (150, 200, 3) uint8 RGB static image of shape. (150, 200, 3). Subsampled from (200,200, 3) image.
steps/observation/robot_obs Tensor (15,) float32 EE position (3), EE orientation in euler angles (3), gripper width (1), joint positions (7), gripper action (1)
steps/observation/structured_language_instruction Tensor string One of 25 possible structured language instructions, see list in https://arxiv.org/pdf/2210.01911.pdf Table 2.
steps/reward Scalar float32
  • Citation:
@inproceedings{rosete2022tacorl,
author = {Erick Rosete-Beas and Oier Mees and Gabriel Kalweit and Joschka Boedecker and Wolfram Burgard},
title = {Latent Plans for Task Agnostic Offline Reinforcement Learning},
journal = {Proceedings of the 6th Conference on Robot Learning (CoRL)},
year = {2022}
}
@inproceedings{mees23hulc2,
title={Grounding  Language  with  Visual  Affordances  over  Unstructured  Data},
author={Oier Mees and Jessica Borja-Diaz and Wolfram Burgard},
booktitle = {Proceedings of the IEEE International Conference on Robotics and Automation (ICRA)},
year={2023},
address = {London, UK}
}