Gridenvs is an open-source library that allows to easily create gym gridworld environments. To test some examples, run keyboard_agent.py.
You can easily create an environment by implementing the following three functions:
- get_init_state(): returns a dictionary containing the internal state of the environment. The state dict will contain all the necessary information of the game, and has to come with at least two elements: "world" (the GridWorld object) and "hero" (the agent GridObject).
- get_next_state(state, action): implements the logic of the environment, and returns a tuple
(state, reward, end_of_episode, info)just as gym's env.step() function, except for the observation, that is automatically generated from the returned state. - get_gridstate(state): returns a tuple consisting of the grid size and a list of objects to be rendered. This is used to generate the observation from the state.
Most environments consist of a single agent that moves in a grid (i.e. action affect a single object). If this is your case, consider inheriting from the HeroEnv class. To easily create a grid map, we provide a function that takes as input a list of strings, such as:
["WWWWWWWWWW",
"WD.W....KW",
"W..W.....W",
"W..W..WWWW",
"W........W",
"W........W",
"W..WWWW..W",
"W........W",
"WH.......W",
"WWWWWWWWWW"]
See, for instance, the key-door examples.
To be able to create your environment with gym.make() you will need to register it. See gym's creating environments guide or check examples/__init__.py.
For other types of games, consider deriving directly from the Env class.
Clone the repository:
git clone https://github.com/aig-upf/gridenvs.git
cd gridenvs
(Optional) If you want to create a virtual environment, you can do it by:
python3 -m venv my_venv
source my_venv/bin/activate
Install gridenvs:
pip3 install -r requirements.txt
When using it in other projects, make sure gridenvs is in your $PYTHONPATH (e.g. by adding it with export PYTHONPATH=\my\path\to\gridenvs).
If you are using an environment from the examples package, import it so that environments get registered to gym (i.e. import gridenvs.examples).