Humanoid Maze Environment. In this task, immovable blocks are placed to confine the agent to a U-shaped corridor. That is, blocks are placed

Create a flow-specific environment, as provided by this repository. The original environments are wrapped by an additional environment that a

Ant Maze Environment. In this task, immovable blocks are placed to confine the agent to a U-shaped corridor. That is, blocks are placed every

RL algorithm class. Supports the training of TD3, SAC, and PPO policies. Attributes ---------- policy : type [ hbaselines.base_polic

Non-flow variant of the ring road environment. Attributes ---------- initial_state : str or None the initial state. Must be one o

Ant Four Rooms Environment. In this environment, an agent is placed in a four-room network whose structure is represented in the figure below

Visual Ant Maze Environment. In this task, immovable blocks are placed to confine the agent to a U-shaped corridor. That is, blocks are place

Base Policy. Attributes ---------- sess : tf.compat.v1.Session the current TensorFlow session ob_space : gym.spaces.*

Ant Fall Environment. In this task, the agent is initialized on a platform of height 4. Immovable blocks are placed everywhere except at (-8,

Ant Push Environment. In this task, immovable blocks are placed every where except at (0,0), (-8,0), (-8,8), (0,8), (8,8), (16,8), and (0,16)

Experience replay buffer for independent multi-agent settings. This replay buffer supports centralized training by including a full-states te

Pendulum environment class. In this environment, an inverted pendulum object is tasked with reaching an end goal consisting of the desired jo

Experience replay buffer.

Experience replay buffer for shared multi-agent settings. This replay buffer supports centralized training by including a full-states term fo

UR5 environment class. In this environment, a UR5 reacher object is tasked with reaching an end goal consisting of the desired joint position

Ant Gather environment. See: https://arxiv.org/pdf/1704.03012.pdf

Dummy MuJoCo data object. Used for testing purposes.

Hierarchical variant of ReplayBuffer. Attributes ---------- buffer_size : int Max number of transitions to store in the buffer. W

SAC-compatible multi-agent feedforward neural. The attributes described in this docstring are only used if the `maddpg` parameter is set to T

A little 2D point whose life goal is to reach a target. Attributes ---------- render_dt_msec : float seconds before the next fram

Multi-agent variant of the ring environment.

Snake Gather environment.

Swimmer Gather environment.

Closed network variant of AVEnv.

Closed network variant of AVMultiAgentEnv.

Environment for training automated vehicles in a mixed-autonomy setting. Required from env_params: * max_accel: scaling factor for the AV ac

Multi-agent variants of AVEnv.

Open network variant of AVEnv.

Open network variant of AVMultiAgentEnv.

Bipedal Obstacles environment. In this environment, a bipedal agent is placed in an open field with obstacles scattered throughout the world.

Bipedal Soccer environment. In this environment, a bipedal agent is placed in an open field with a soccer ball. The agent is rewarded for mov

Get the parameters as a flat vector.

Humanoid Four Rooms Environment. In this environment, an agent is placed in a four-room network whose structure is represented in the figure

Humanoid Push Environment. In this task, immovable blocks are placed every where except at (0,0), (-8,0), (-8,8), (0,8), (8,8), (16,8), and (

Single agent variant of the ring environment.

Environment sampler object. Attributes ---------- env : gym.Env the training / evaluation environment

Set the parameters from a flat vector.

Logger object for evaluation trajectory data. This method logs, save, and plots trajectory data during evaluations for a number of tasks.

Gym representation of the Ant MuJoCo environment.

Gym representation of AntMaze, AntPush, or AntFall.

Custom installation procedure.

Dummy MuJoCo model object. Used for testing purposes.

Base environment class. Supports the UR5 and Pendulum environments from: Levy, Andrew, et al. "Learning Multi-Level Hierarchies with Hindsig

Feed-forward neural network actor-critic policy. Attributes ---------- sess : tf.compat.v1.Session the current TensorFlow session

SAC-compatible feedforward policy. Attributes ---------- sess : tf.compat.v1.Session the current TensorFlow session ob_space

Feed-forward neural network policy. Attributes ---------- learning_rate : float the learning rate n_minibatches : int

TRPO-compatible feedforward policy.. Attributes ---------- gamma : float the discount factor lam : float factor for t

Goal-conditioned hierarchical reinforcement learning model. Attributes ---------- num_levels : int number of levels within the hi

TD3-compatible goal-conditioned hierarchical policy.

SAC-compatible goal-conditioned hierarchical policy.

Humanoid mujoco environment.

Humanoid mujoco environment.

Humanoid Fall Environment. In this task, the agent is initialized on a platform of height 4. Immovable blocks are placed everywhere except at

Gym representation of HumanoidMaze, HumanoidPush, or HumanoidFall.

Visual Humanoid Maze Environment. In this task, immovable blocks are placed to confine the agent to a U-shaped corridor. That is, blocks are

Dummy MuJoCo simulator object. Used for testing purposes.

Dummy MuJoCo viewer object. Used for testing purposes.

Movable attributes.

Multi-agent base policy. This policy supports training off-policy variants of three popular multi-agent algorithms: * Independent learne

TD3-compatible multi-agent feedforward neural. The attributes described in this docstring are only used if the `maddpg` parameter is set to T

PPO-compatible multi-agent feedforward neural network policy.

TRPO-compatible multi-agent feedforward neural network policy.

See parent class.

See parent class.

TD3-compatible multi-agent goal-conditioned hierarchical policy.

SAC-compatible multi-agent goal-conditioned hierarchical policy.

See parent class.

Ray-compatible variant of the environment sampler object. Used to collect samples in parallel.

Snake Mujoco environment.

Tests the automated vehicles single agent environments.

Tests the automated vehicles multi-agent environments.

Test GoalConditionedPolicy in hbaselines/goal_conditioned/base.py.

Test the Ant* environments in envs/efficient_hrl.

Test the Humanoid* environments in envs/efficient_hrl/.

Test the environment utility methods.

Unit tests for the classes and methods in utils/eval.py.

Tests the runner scripts in the experiments folder.

Test the environments in envs/hac/.

Tests for the HierReplayBuffer object.

Test the functionality of each of the trainable mixed-autonomy envs. Each of these environments are tests for the following cases: 1. the ob

Test MultiAgentPolicy in hbaselines/multiagent/base.py.

Tests for the MultiReplayBuffer object.

Test FeedForwardPolicy in hbaselines/fcnet/ppo.py.

Test MultiFeedForwardPolicy in hbaselines/multiagent/ppo.py.

Tests the Pendulum environment class.

Tests for the experiments/plot.py script.

Test the functionality of features in envs/point2d.py.

Test Policy in hbaselines/base_policies/policy.py.

Test the components of the RLAlgorithm algorithm.

Tests for the ReplayBuffer object.

Test the reward_fns method.

Test the functionality of features in ring_nonflow.py.

Test FeedForwardPolicy in hbaselines/fcnet/sac.py.

Test GoalConditionedPolicy in hbaselines/goal_conditioned/sac.py.

Test MultiFeedForwardPolicy in hbaselines/multiagent/sac.py.

Test MultiFeedForwardPolicy in hbaselines/multiagent/h_sac.py.

Tests for the environments in envs/snn4hrl.

Tests for the SharedReplayBuffer object.

Test FeedForwardPolicy in hbaselines/fcnet/td3.py.

Test GoalConditionedPolicy in hbaselines/goal_conditioned/td3.py.