rlu_control_suite

  • Description:

RL Unplugged is suite of benchmarks for offline reinforcement learning. The RL Unplugged is designed around the following considerations: to facilitate ease of use, we provide the datasets with a unified API which makes it easy for the practitioner to work with all data in the suite once a general pipeline has been established.

The datasets follow the RLDS format to represent steps and episodes.

DeepMind Control Suite Tassa et al., 2018 is a set of control tasks implemented in MuJoCo Todorov et al., 2012. We consider a subset of the tasks provided in the suite that cover a wide range of difficulties.

Most of the datasets in this domain are generated using D4PG. For the environments Manipulator insert ball and Manipulator insert peg we use V-MPO Song et al., 2020 to generate the data as D4PG is unable to solve these tasks. We release datasets for 9 control suite tasks. For details on how the dataset was generated, please refer to the paper.

DeepMind Control Suite is a traditional continuous action RL benchmark. In particular, we recommend you test your approach in DeepMind Control Suite if you are interested in comparing against other state of the art offline RL methods.

@inproceedings{gulcehre2020rl,
 title = {RL Unplugged: A Suite of Benchmarks for Offline Reinforcement Learning},
 author = {Gulcehre, Caglar and Wang, Ziyu and Novikov, Alexander and Paine, Thomas and G'{o}mez, Sergio and Zolna, Konrad and Agarwal, Rishabh and Merel, Josh S and Mankowitz, Daniel J and Paduraru, Cosmin and Dulac-Arnold, Gabriel and Li, Jerry and Norouzi, Mohammad and Hoffman, Matthew and Heess, Nicolas and de Freitas, Nando},
 booktitle = {Advances in Neural Information Processing Systems},
 pages = {7248--7259},
 volume = {33},
 year = {2020}
}

rlu_control_suite/cartpole_swingup (default config)

  • Dataset size: 2.12 MiB

  • Auto-cached (documentation): Yes

  • Splits:

Split Examples
'train' 40
  • Feature structure:
FeaturesDict({
    'episode_id': int64,
    'steps': Dataset({
        'action': Tensor(shape=(1,), dtype=float32),
        'discount': float32,
        'is_first': bool,
        'is_last': bool,
        'is_terminal': bool,
        'observation': FeaturesDict({
            'position': Tensor(shape=(3,), dtype=float32),
            'velocity': Tensor(shape=(2,), dtype=float32),
        }),
        'reward': float32,
    }),
    'timestamp': int64,
})
  • Feature documentation:
Feature Class Shape Dtype Description
FeaturesDict
episode_id Tensor int64
steps Dataset
steps/action Tensor (1,) float32
steps/discount Tensor float32
steps/is_first Tensor bool
steps/is_last Tensor bool
steps/is_terminal Tensor bool
steps/observation FeaturesDict
steps/observation/position Tensor (3,) float32
steps/observation/velocity Tensor (2,) float32
steps/reward Tensor float32
timestamp Tensor int64

rlu_control_suite/cheetah_run

  • Dataset size: 36.58 MiB

  • Auto-cached (documentation): Yes

  • Splits:

Split Examples
'train' 300
  • Feature structure:
FeaturesDict({
    'episode_id': int64,
    'steps': Dataset({
        'action': Tensor(shape=(6,), dtype=float32),
        'discount': float32,
        'is_first': bool,
        'is_last': bool,
        'is_terminal': bool,
        'observation': FeaturesDict({
            'position': Tensor(shape=(8,), dtype=float32),
            'velocity': Tensor(shape=(9,), dtype=float32),
        }),
        'reward': float32,
    }),
    'timestamp': int64,
})
  • Feature documentation:
Feature Class Shape Dtype Description
FeaturesDict
episode_id Tensor int64
steps Dataset
steps/action Tensor (6,) float32
steps/discount Tensor float32
steps/is_first Tensor bool
steps/is_last Tensor bool
steps/is_terminal Tensor bool
steps/observation FeaturesDict
steps/observation/position Tensor (8,) float32
steps/observation/velocity Tensor (9,) float32
steps/reward Tensor float32
timestamp Tensor int64

rlu_control_suite/finger_turn_hard

  • Dataset size: 47.61 MiB

  • Auto-cached (documentation): Yes

  • Splits:

Split Examples
'train' 500
  • Feature structure:
FeaturesDict({
    'episode_id': int64,
    'steps': Dataset({
        'action': Tensor(shape=(2,), dtype=float32),
        'discount': float32,
        'is_first': bool,
        'is_last': bool,
        'is_terminal': bool,
        'observation': FeaturesDict({
            'dist_to_target': Tensor(shape=(1,), dtype=float32),
            'position': Tensor(shape=(4,), dtype=float32),
            'target_position': Tensor(shape=(2,), dtype=float32),
            'velocity': Tensor(shape=(3,), dtype=float32),
        }),
        'reward': float32,
    }),
    'timestamp': int64,
})
  • Feature documentation:
Feature Class Shape Dtype Description
FeaturesDict
episode_id Tensor int64
steps Dataset
steps/action Tensor (2,) float32
steps/discount Tensor float32
steps/is_first Tensor bool
steps/is_last Tensor bool
steps/is_terminal Tensor bool
steps/observation FeaturesDict
steps/observation/dist_to_target Tensor (1,) float32
steps/observation/position Tensor (4,) float32
steps/observation/target_position Tensor (2,) float32
steps/observation/velocity Tensor (3,) float32
steps/reward Tensor float32
timestamp Tensor int64

rlu_control_suite/fish_swim

  • Dataset size: 32.81 MiB

  • Auto-cached (documentation): Yes

  • Splits:

Split Examples
'train' 200
  • Feature structure:
FeaturesDict({
    'episode_id': int64,
    'steps': Dataset({
        'action': Tensor(shape=(5,), dtype=float32),
        'discount': float32,
        'is_first': bool,
        'is_last': bool,
        'is_terminal': bool,
        'observation': FeaturesDict({
            'joint_angles': Tensor(shape=(7,), dtype=float32),
            'target': Tensor(shape=(3,), dtype=float32),
            'upright': Tensor(shape=(1,), dtype=float32),
            'velocity': Tensor(shape=(13,), dtype=float32),
        }),
        'reward': float32,
    }),
    'timestamp': int64,
})
  • Feature documentation:
Feature Class Shape Dtype Description
FeaturesDict
episode_id Tensor int64
steps Dataset
steps/action Tensor (5,) float32
steps/discount Tensor float32
steps/is_first Tensor bool
steps/is_last Tensor bool
steps/is_terminal Tensor bool
steps/observation FeaturesDict
steps/observation/joint_angles Tensor (7,) float32
steps/observation/target Tensor (3,) float32
steps/observation/upright Tensor (1,) float32
steps/observation/velocity Tensor (13,) float32
steps/reward Tensor float32
timestamp Tensor int64

rlu_control_suite/humanoid_run

Split Examples
'train' 3,000
  • Feature structure:
FeaturesDict({
    'episode_id': int64,
    'steps': Dataset({
        'action': Tensor(shape=(21,), dtype=float32),
        'discount': float32,
        'is_first': bool,
        'is_last': bool,
        'is_terminal': bool,
        'observation': FeaturesDict({
            'com_velocity': Tensor(shape=(3,), dtype=float32),
            'extremities': Tensor(shape=(12,), dtype=float32),
            'head_height': Tensor(shape=(1,), dtype=float32),
            'joint_angles': Tensor(shape=(21,), dtype=float32),
            'torso_vertical': Tensor(shape=(3,), dtype=float32),
            'velocity': Tensor(shape=(27,), dtype=float32),
        }),
        'reward': float32,
    }),
    'timestamp': int64,
})
  • Feature documentation:
Feature Class Shape Dtype Description
FeaturesDict
episode_id Tensor int64
steps Dataset
steps/action Tensor (21,) float32
steps/discount Tensor float32
steps/is_first Tensor bool
steps/is_last Tensor bool
steps/is_terminal Tensor bool
steps/observation FeaturesDict
steps/observation/com_velocity Tensor (3,) float32
steps/observation/extremities Tensor (12,) float32
steps/observation/head_height Tensor (1,) float32
steps/observation/joint_angles Tensor (21,) float32
steps/observation/torso_vertical Tensor (3,) float32
steps/observation/velocity Tensor (27,) float32
steps/reward Tensor float32
timestamp Tensor int64

rlu_control_suite/manipulator_insert_ball

  • Dataset size: 385.41 MiB

  • Auto-cached (documentation): No

  • Splits:

Split Examples
'train' 1,500
  • Feature structure:
FeaturesDict({
    'episode_id': int64,
    'steps': Dataset({
        'action': Tensor(shape=(5,), dtype=float32),
        'discount': float32,
        'is_first': bool,
        'is_last': bool,
        'is_terminal': bool,
        'observation': FeaturesDict({
            'arm_pos': Tensor(shape=(16,), dtype=float32),
            'arm_vel': Tensor(shape=(8,), dtype=float32),
            'hand_pos': Tensor(shape=(4,), dtype=float32),
            'object_pos': Tensor(shape=(4,), dtype=float32),
            'object_vel': Tensor(shape=(3,), dtype=float32),
            'target_pos': Tensor(shape=(4,), dtype=float32),
            'touch': Tensor(shape=(5,), dtype=float32),
        }),
        'reward': float32,
    }),
    'timestamp': int64,
})
  • Feature documentation:
Feature Class Shape Dtype Description
FeaturesDict
episode_id Tensor int64
steps Dataset
steps/action Tensor (5,) float32
steps/discount Tensor float32
steps/is_first Tensor bool
steps/is_last Tensor bool
steps/is_terminal Tensor bool
steps/observation FeaturesDict
steps/observation/arm_pos Tensor (16,) float32
steps/observation/arm_vel Tensor (8,) float32
steps/observation/hand_pos Tensor (4,) float32
steps/observation/object_pos Tensor (4,) float32
steps/observation/object_vel Tensor (3,) float32
steps/observation/target_pos Tensor (4,) float32
steps/observation/touch Tensor (5,) float32
steps/reward Tensor float32
timestamp Tensor int64

rlu_control_suite/manipulator_insert_peg

  • Dataset size: 385.73 MiB

  • Auto-cached (documentation): No

  • Splits:

Split Examples
'train' 1,500
  • Feature structure:
FeaturesDict({
    'episode_id': int64,
    'steps': Dataset({
        'action': Tensor(shape=(5,), dtype=float32),
        'discount': float32,
        'is_first': bool,
        'is_last': bool,
        'is_terminal': bool,
        'observation': FeaturesDict({
            'arm_pos': Tensor(shape=(16,), dtype=float32),
            'arm_vel': Tensor(shape=(8,), dtype=float32),
            'hand_pos': Tensor(shape=(4,), dtype=float32),
            'object_pos': Tensor(shape=(4,), dtype=float32),
            'object_vel': Tensor(shape=(3,), dtype=float32),
            'target_pos': Tensor(shape=(4,), dtype=float32),
            'touch': Tensor(shape=(5,), dtype=float32),
        }),
        'reward': float32,
    }),
    'timestamp': int64,
})
  • Feature documentation:
Feature Class Shape Dtype Description
FeaturesDict
episode_id Tensor int64
steps Dataset
steps/action Tensor (5,) float32
steps/discount Tensor float32
steps/is_first Tensor bool
steps/is_last Tensor bool
steps/is_terminal Tensor bool
steps/observation FeaturesDict
steps/observation/arm_pos Tensor (16,) float32
steps/observation/arm_vel Tensor (8,) float32
steps/observation/hand_pos Tensor (4,) float32
steps/observation/object_pos Tensor (4,) float32
steps/observation/object_vel Tensor (3,) float32
steps/observation/target_pos Tensor (4,) float32
steps/observation/touch Tensor (5,) float32
steps/reward Tensor float32
timestamp Tensor int64

rlu_control_suite/walker_stand

  • Dataset size: 31.78 MiB

  • Auto-cached (documentation): Yes

  • Splits:

Split Examples
'train' 200
  • Feature structure:
FeaturesDict({
    'episode_id': int64,
    'steps': Dataset({
        'action': Tensor(shape=(6,), dtype=float32),
        'discount': float32,
        'is_first': bool,
        'is_last': bool,
        'is_terminal': bool,
        'observation': FeaturesDict({
            'height': Tensor(shape=(1,), dtype=float32),
            'orientations': Tensor(shape=(14,), dtype=float32),
            'velocity': Tensor(shape=(9,), dtype=float32),
        }),
        'reward': float32,
    }),
    'timestamp': int64,
})
  • Feature documentation:
Feature Class Shape Dtype Description
FeaturesDict
episode_id Tensor int64
steps Dataset
steps/action Tensor (6,) float32
steps/discount Tensor float32
steps/is_first Tensor bool
steps/is_last Tensor bool
steps/is_terminal Tensor bool
steps/observation FeaturesDict
steps/observation/height Tensor (1,) float32
steps/observation/orientations Tensor (14,) float32
steps/observation/velocity Tensor (9,) float32
steps/reward Tensor float32
timestamp Tensor int64

rlu_control_suite/walker_walk

  • Dataset size: 31.78 MiB

  • Auto-cached (documentation): Yes

  • Splits:

Split Examples
'train' 200
  • Feature structure:
FeaturesDict({
    'episode_id': int64,
    'steps': Dataset({
        'action': Tensor(shape=(6,), dtype=float32),
        'discount': float32,
        'is_first': bool,
        'is_last': bool,
        'is_terminal': bool,
        'observation': FeaturesDict({
            'height': Tensor(shape=(1,), dtype=float32),
            'orientations': Tensor(shape=(14,), dtype=float32),
            'velocity': Tensor(shape=(9,), dtype=float32),
        }),
        'reward': float32,
    }),
    'timestamp': int64,
})
  • Feature documentation:
Feature Class Shape Dtype Description
FeaturesDict
episode_id Tensor int64
steps Dataset
steps/action Tensor (6,) float32
steps/discount Tensor float32
steps/is_first Tensor bool
steps/is_last Tensor bool
steps/is_terminal Tensor bool
steps/observation FeaturesDict
steps/observation/height Tensor (1,) float32
steps/observation/orientations Tensor (14,) float32
steps/observation/velocity Tensor (9,) float32
steps/reward Tensor float32
timestamp Tensor int64