reubenwong97 / NFSP-PEG-GridWorld

Implementation of Neural Fictitious Self-Play for a GridWorld based Multi-Agent Pursuit Evasion Game

Epsilon-greedy Neural Fictitious Self-Play in Pursuit Evasion Grid-World

PEG Environment

Credits to zuoxingdong for the base mazelab environment.

Mazelab supports maze generation and game playing API for single player environments out of the box. I modified the environment to support multi-agent gameplay.

NFSP

This repo includes my implementation of the NFSP algorithm. Provided in the main.py file is the driver code for interacting with my environment, as well as code for logging the experiment.

TODO

• Improve logging
• Improve hyperparameter management