Module: tf_agents.environments

Environments module.

Modules

batched_py_environment module: Treat multiple non-batch environments as a single batch environment.

gym_wrapper module: Wrapper providing a PyEnvironmentBase adapter for Gym environments.

parallel_py_environment module: Runs multiple environments in parallel processes and steps them in batch.

py_environment module: Python RL Environment API.

random_py_environment module: Environment implementation that generates random observations.

random_tf_environment module: Utility environment that creates random observations.

suite_gym module: Suite for loading Gym Environments.

tf_environment module: TensorFlow RL Environment API.

tf_py_environment module: Wrapper for PyEnvironments into TFEnvironments.

trajectory_replay module: A Driver-like object that replays Trajectories.

utils module: Common utilities for TF-Agents Environments.

wrappers module: Environment wrappers.

Classes

class ActionClipWrapper: Wraps an environment and clips actions to spec before applying.

class ActionDiscretizeWrapper: Wraps an environment with continuous actions and discretizes them.

class ActionOffsetWrapper: Offsets actions to be zero-based.

class ActionRepeat: Repeates actions over n-steps while acummulating the received reward.

class BatchedPyEnvironment: Batch together multiple py environments and act as a single batch.

class FlattenObservationsWrapper: Wraps an environment and flattens nested multi-dimensional observations.

class GoalReplayEnvWrapper: Adds a goal to the observation, used for HER (Hindsight Experience Replay).

class HistoryWrapper: Adds observation and action history to the environment's observations.

class ObservationFilterWrapper: Filters observations based on an array of indexes.

class OneHotActionWrapper: Converts discrete action to one_hot format.

class ParallelPyEnvironment: Batch together environments and simulate them in external processes.

class PerformanceProfiler: End episodes after specified number of steps.

class PyEnvironment: Abstract base class for Python RL environments.

class PyEnvironmentBaseWrapper: PyEnvironment wrapper forwards calls to the given environment.

class RandomPyEnvironment: Randomly generates observations following the given observation_spec.

class RandomTFEnvironment: Randomly generates observations following the given observation_spec.

class RunStats: Wrapper that accumulates run statistics as the environment iterates.

class TFEnvironment: Abstract base class for TF RL environments.

class TFPyEnvironment: Exposes a Python environment as an in-graph TF environment.

class TimeLimit: End episodes after specified number of steps.

class TrajectoryReplay: A helper that replays a policy against given Trajectory observations.

Functions

validate_py_environment(...): Validates the environment follows the defined specs.