Hang Chen, Ruiqi Wang

This project is marked finished if not updated for the future work. A final project report is also uploaded to this repo. A demo video is available here - https://youtu.be/X_JOLY-A14g

This repository is a final project for CISC889 Multi-Agent Reinforcement Learning (MARL). Base on the video game Mr. Pac-Man framework from CS188 UCB, we are going to design a strategy to apply multiple Pacman agents to eat pellets in the maze. To get a higher score, Pacman should eat all pellets as quickly as possible while avoid being eaten by the ghosts.

• (2 weeks) implement SARSA on 1 ghosts and 1 pacman;
• (1 weeks) extend to 1 ghosts and 2 pacmans; (ended up to several pacmans)
• (3 weeks) figure out and implement proper MARL algorithm to replace SARSA; (ended up adding a DQN agent)