Engine: an environment-building tool for safe exploration research. The Engine() class entails everything to do with the tasks and safety

Base used to allow for convenient hierarchies of environments

A dictionnary of lists of same size. Dictionnary items can be accessed using `.` notation and list items using `[]` notation. Example:

A concurrent execution of environments in multiple processes.

Raised when we fail to sample a valid distribution of objects or goals

Simple utility class for getting mujoco-specific info about a robot

A mapping from tokens to ids with a capacity of `max_size` words. It can be saved in a `vocab.json` file.

This environment is a grid with randomly located letters on it We ensure that there is a clean path to any of the letters (a path that includ

This agent allows user to play with Safety's Point agent. Use the UP and DOWN arrows to move forward and back and use '<' and '>' to rot

This agent picks actions randomly

The Advantage Actor-Critic algorithm.

An environment where a myopic agent will fail. The two possible goals are "Reach blue then green" or "Reach blue then red".

An agent. It is able: - to choose an action given an observation, - to analyze the feedback (i.e. reward and done state) of its action.

The base class for RL algorithms.

A simple wrapper for a gym-minigrid environment. This implements propositions on top of the minigrid.

The Proximal Policy Optimization algorithm ([Schulman et al., 2015](https://arxiv.org/abs/1707.06347)).

This environment is a modification of the Safety-Gym's environment. There is no "goal circle" but rather a collection of zones that the a