package rlds_creator

Get desktop application:
View/edit binary Protocol Buffers messages

Specification of an environment that users interact with. Next ID: 16

Used in: StudySpec, client.EpisodeMetadata, client.SelectEnvironmentResponse

• optional string id = 7

  Unique ID of the environment in the study.

• optional string name = 1

  Human readable name of the environment.

• optional string additional_instructions = 2

  Additional instructions for the environment. When a study contains several environments, you can use this field to provide more specific instructions to the user for this environment.

• optional bool sync = 3

  Tf true, then the environment transitions to the next state after an action (e.g. pressing a key). Otherwise, state transitions will happen at a fixed rate, e.g. 15 steps per second.

• optional int32 max_episode_steps = 10

  Maximum number of steps in each episode. If not specified, this will depend on the settings of the environments.

• oneof type

  Type of the environment and its settings.

  • EnvironmentSpec.Procgen procgen = 4

  • EnvironmentSpec.Atari atari = 5

  • EnvironmentSpec.DMLab dmlab = 8

  • EnvironmentSpec.NetHack net_hack = 13

  • EnvironmentSpec.Robosuite robosuite = 11

  • EnvironmentSpec.RoboDesk robodesk = 12

Settings for the Atari environment. See third_party/py/atari_py.

Used in: EnvironmentSpec

• optional string id = 1

  ID of the Atari environment, e.g. MontezumaRevenge.

• optional bool sticky_actions = 2

  When sticky actions are enabled, the environment repeats the previous action with probability 0.25, instead of playing the action given by the agent.

Settings for the DMLab environments. See third_party/labyrinth/.

Used in: EnvironmentSpec

• optional string id = 1

  ID of the DMLab environment, i.e. level, e.g. explore_goal_locations_small.

Settings for the NetHack environments. See github.com/facebookresearch/nle.

Used in: EnvironmentSpec

• optional string id = 1

  ID of the NetHack environment, e.g. NetHackScore-v0.

Settings for the Procgen environment. See third_party/py/procgen.

Used in: EnvironmentSpec

• optional string id = 1

  ID of the Procgen environment, e.g. maze.

• optional Procgen.DistributionMode distribution_mode = 2

• optional int32 num_levels = 3

  Number of unique levels to generate or 0 to use unlimited levels.

• optional int32 start_level = 4

  Start level.

• optional int32 level_repeats = 5

  Number of levels before resetting the environment.

• optional bool use_sequential_levels = 6

  If true, then levels will advance sequentially.

• optional int32 rand_seed = 7

  Seed for random number generation. If not specified, then a random one will be used.

Distribution mode; see DISTRUBUTION_MODE_DICT in procgen/env.py.

Used in: Procgen

• DISTRIBUTION_UNSPECIFIED = 0

• DISTRIBUTION_EASY = 1

• DISTRIBUTION_HARD = 2

• DISTRIBUTION_EXTREME = 3

• DISTRIBUTION_MEMORY = 4

• DISTRIBUTION_EXPLORATION = 5

  In exploration mode, number of levels and starting level parameters are ignored.

Settings for the RoboDesk environment. See https://github.com/google-research/robodesk.

Used in: EnvironmentSpec

• optional string task = 1

  Name of the task, e.g. open_slide. See config.py for the possible options.

• optional RoboDesk.Reward reward = 2

• optional int32 action_repeat = 3

  Number of times to repeat each action.

• optional int32 image_size = 4

  Size of the images in the observations. This would be different from the image renderings that are displayed to the users when collecting the trajectories.

Type of the reward.

Used in: RoboDesk

• REWARD_UNSPECIFIED = 0

• REWARD_DENSE = 1

• REWARD_SPARSE = 2

• REWARD_SUCCESS = 3

Settings for the Robosuite environment. See third_party/py/robosuite.

Used in: EnvironmentSpec

• optional string id = 1

  ID of the Robosuite environment, e.g. Lift.

• repeated string robots = 2

  Robot(s) to use in the environment, e.g. Panda.

• optional string config = 3

  Environment configuration, e.g. single-arm-opposed.

• optional string arm = 4

  Arm to control, e.g. right or left.

• repeated string cameras = 10

  Cameras to use, e.g. agentview or frontview. If empty, then agentview camera will be used.

• optional string controller = 6

  Controller to use, either IK_POSE or OSC_POSE.

• optional float pos_sensitivity = 7

  Position scaling for user inputs.

• optional float rot_sensitivity = 8

  Rotation scaling for user inputs.

• optional bool use_camera_obs = 9

  If true, then observations will include rendered images.

A study contains a set of episodes provided by the users, i.e. recorded as they interact with the environments in the study. Next ID: 17

Used in: client.EpisodeMetadata

• optional string id = 1

  Unique ID of the episode.

• optional string study_id = 2

  Unique ID of the study that the episode is collected for.

• optional string environment_id = 9

  Unique ID of the environment in the study specification.

• optional User user = 3

  User that provided the episode.

• optional string session_id = 4

  Unique ID of the session that the episode is part of. Episodes in a session will be from the same user. A user may have multiple sessions.

• optional Episode.State state = 5

• optional google.protobuf.Timestamp start_time = 6

  Starting and end timestamps.

• optional google.protobuf.Timestamp end_time = 7

• optional google.protobuf.Struct metadata = 8

  Metadata about the episode, e.g. parameters of the environment.

• optional int32 num_steps = 10

  Number of steps.

• optional float total_reward = 12

  Total reward over all steps.

• optional Episode.Storage storage = 11

• repeated Tag tags = 13

  A list of tags for the episode.

• optional string notes = 14

  Notes about the episode.

• optional string controller_id = 15

  ID of the input (e.g. gamepad) controller used by the user, excluding the keyboard (which is always active and can be used together with the controller). This will be the last controller used for an action in the episode in the unlikely case of multiple controllers.

• map<int32, StepMetadata> step_metadata = 16

  Metadata about the steps of the epsiode keyed by their 0-based index.

State of the episode.

Used in: Episode

• STATE_UNSPECIFIED = 0

• STATE_COMPLETED = 1

  Episode completed successfully and confirmed by the user.

• STATE_ABANDONED = 2

  Episode is abandoned by the user, e.g. user stopped responding.

• STATE_CANCELLED = 3

  Episode is explicitly cancelled by the user before completion.

• STATE_REJECTED = 4

  Episode completed successfully, but at the end the user decided to reject it.

Describes how the episode data was stored.

Used in: Episode

• oneof type

  • Storage.EnvironmentLogger environment_logger = 1

  • Storage.Pickle pickle = 2

Envlogger. See pypi.org/project/envlogger.

Used in: Storage

• optional string tag_directory = 1

  Tag directory.

• optional int32 index = 2

  Index of the episode in the logged episodes under this tag directory.

Pickle files.

Used in: Storage

• optional string path = 1

  Path of the Pickle file that contains the episode data.

A user session associated with a study.

• optional string id = 1

  Unique ID of the session.

• optional string study_id = 2

  Unique ID of the study that the session is associated with.

• optional User user = 3

  User of the session.

• optional Session.State state = 4

• optional google.protobuf.Timestamp start_time = 5

  Starting and end timestamps.

• optional google.protobuf.Timestamp end_time = 6

State of the session.

Used in: Session

• STATE_UNSPECIFIED = 0

• STATE_VALID = 1

  Session is valid.

• STATE_INVALID = 2

  Session is invalid and the corresponding episodes will be ignored.

Metadata about a step of an episode.

Used in: Episode

• repeated Tag tags = 1

  A list of tags for the step.

Specification of a study. Next ID: 9

Used in: StudySpecs, client.EpisodeMetadata, client.SaveStudyRequest, client.SelectStudyResponse, client.SetStudiesResponse

• optional string id = 1

  Unique ID of the study.

• optional string name = 2

  Name of the study, e.g. "Atari games".

• optional string description = 3

  (Short) description of the study.

• optional string instructions = 4

  Instructions for the study. This is shown to the users and should describe the goal of the study and what to do.

• optional User creator = 5

  User who created the study.

• optional google.protobuf.Timestamp creation_time = 8

  Creation timestamp.

• optional StudySpec.State state = 6

• repeated EnvironmentSpec environment_specs = 7

  Specifications of the environments in the study. TODO(sertan): We need to define how the environments will be assigned to users, e.g. users may select a list or one of the environments can be chosen randomly.

State of the study.

Used in: StudySpec

• STATE_UNSPECIFIED = 0

• STATE_ENABLED = 1

  Study is enabled. Users can interact with the environments in the study and provide data.

• STATE_DISABLED = 2

  Study is disabled. Users cannot provide data for the study.

A list of study specifications.

• repeated StudySpec study_specs = 1

Denotes a tag with a label.

Used in: Episode, StepMetadata

• optional string label = 1

  Label of the tag.

Information about a user.

Used in: Episode, Session, StudySpec

• optional string email = 1

  Email of the user.