Skip to content

Reinforcement learning agent that plays Ms. Pac-Man (pretty well)

License

Notifications You must be signed in to change notification settings

anassinator/ms-pacman

Repository files navigation

Ms. PacMan Agent

This is a reinforcement learning agent that plays Ms. PacMan through function approximation.

This is a submission to the third assignment of McGill University's ECSE 526 - Artificial Intelligence course. Details can be found here.

This agent has been able to score an average of over 4000 points each episode, with a maximum recorded score of approximately 12500.

screen shot 2016-11-14 at 13 29 37

Setup

To run this, all one needs is Python 2.7 or above, the Arcade Learning Environment with its Python bindings installed and opencv-python.

If you're running this on OSX, you will also need pygame in order to display the game screen.

Running

To run, simply run:

python play.py

or

python play.py --help

for advanced options.

When running, you should see a window as pictured above (titled ALE Viz). The map and sliced map windows would appear when running with the --map-display option. The map window shows the reduced approximation of the game field, whereas the sliced map window is the portion of the map the agent is currently analyzing. A legend of the colors used in the maps is shown below:

Color Meaning
Blue Clear path
Orange Wall
White Pellet
Cyan Power-up
Magenta Fruit
Red Bad ghost
Green Edible ghost
Yellow Ms. PacMan

About

Reinforcement learning agent that plays Ms. Pac-Man (pretty well)

Topics

Resources

License

Stars

Watchers

Forks

Releases

No releases published

Packages

No packages published

Languages