The Pac-Man Projects

Pac-Man Game

Overview

The Pac-Man projects were developed for UC Berkeley's introductory artificial intelligence course, CS 188. They apply an array of AI techniques to playing Pac-Man. However, these projects don't focus on building AI for video games. Instead, they teach foundational AI concepts, such as informed state-space search, probabilistic inference, and reinforcement learning. These concepts underly real-world application areas such as natural language processing, computer vision, and robotics.

We designed these projects with three goals in mind. The projects allow students to visualize the results of the techniques they implement. They also contain code examples and clear directions, but do not force students to wade through undue amounts of scaffolding. Finally, Pac-Man provides a challenging problem environment that demands creative solutions; real-world AI problems are challenging, and Pac-Man is too.

In our course, these projects have boosted enrollment, teaching reviews, and student engagement. The projects have been field-tested, refined, and debugged over multiple semesters at Berkeley. We are now happy to release them to other universities for educational use.

Important Links

Instructor's Guide for Projects
Using the Projects on the edX Platform
Contact Information

Projects Overview

UNIX/Python Tutorial

This short UNIX/Python tutorial introduces students to the Python programming language and the UNIX environment.


Search

Students implement depth-first, breadth-first, uniform cost, and A* search algorithms. These algorithms are used to solve navigation and traveling salesman problems in the Pacman world.


Multi-Agent Search

Classic Pacman is modeled as both an adversarial and a stochastic search problem. Students implement multiagent minimax and expectimax algorithms, as well as designing evaluation functions.


Reinforcement Learning

Students implement model-based and model-free reinforcement learning algorithms, applied to the AIMA textbook's Gridworld, Pacman, and a simulated crawling robot.


Ghostbusters

Probabilistic inference in a hidden Markov model tracks the movement of hidden ghosts in the Pacman world. Students implement exact inference using the forward algorithm and approximate inference via particle filters.


Classification

Students implement standard machine learning classification algorithms using Naive Bayes, Perceptron, and MIRA models to classify digits. Students extend this by implementing a behavioral cloning Pacman agent.


Contest: Pacman Capture the Flag

Students create strategies for a team of two agents to play a multi-player capture-the-flag variant of Pacman.


Technical Notes

The Pac-Man projects are written in pure Python 2.7 and do not depend on any packages external to a standard Python distribution.


Support

This project was supported by the National Science foundation under CAREER grant 0643742. Any opinions, findings and conclusions or recommendations expressed in this material are those of the author(s) and do not necessarily reflect the views of the National Science Foundation (NSF).


Credits

The projects were developed by John DeNero, Dan Klein, Pieter Abbeel, and many others.