PacMan_AI

This is the term project for Artificial Intelligence course. Acknowledgement: The project is built on top of the PacMan Project from UC Berkley’s CS188: Introduction to Artificial Intelligence course. Link: https://inst.eecs.berkeley.edu/~cs188/sp21/projects/

Progress:

1. Modify the game to change difficulty

• Include more ghosts (Up to 8 ghosts)
• Directional Ghosts (Choose move that minimizes mahattan distance)
• Random Ghosts
• Random Ghosts with patterns (tendency to move horizontally and vertically)
• Intelligent Ghosts (Choose move that minimizes true distance with some stochasticity)
• Incorporate different types of ghosts to the game (Right now only all random ghosts or all directional ghosts are possible)
• Implement ghosts that can eat dots

2. Minimax Pacman & Ghosts

• Implement baseline algorithms of Minimax, Expectimax with Alpha-Beta pruning
• Implement Minimax, Expectimax with Stochastic Gradient Descent (SGD) with hand-crafted feature sets for evaluation function

3. Reinforcement Learning

• Approximate Q-learning with hand-crafted feature sets
• Baseline (Double) Deep Q-Learning
• (Double) Deep Q-Learning with actor-critic
• (Double) Deep Q-Learning with dual networks
• (Double) Deep Q-Learning with prioritized replay