BombermanRL

We explore the extent to which RL can be applied to the bomberman problem that is traditionally solved with A* and BFS.

What is new and interesting about this project is not just that we found some interesting results, our game also looks good.

Setup

This project does require at least Python 3.9, probably even 3.10.

The setup is simple:

$ git clone git@github.com:ayushTNM/BombermanRL.git   # assuming SSH is set up
$ cd BombermanRL
$ python3 -m venv venv                                # create a virtual environment
$ source venv/bin/activate                            # activate virtual environment
(venv) $ pip install -r requirements.txt
(venv) $ cd src
(venv) $ python3 menu.py -h

This will show all of the parameters you can change from the CLI.

To run the file with default parameters, just type:

(venv) $ python3 menu.py

Some fundamental parameters to look into:

-o [plotname]        # store the learning curves as a plot
-m                   # flag to turn on menu (if you want to play yourself)

Demo

5x5 with 6 crates

5x5 with 7 crates

8x8 with 6 crates

9x9 with 4 crates

10x10 with 6 crates