Ms Pac-Man is a game released by Midway Manufacturing in 1982. It is based on the original Pac-Man game and adds new maps and semi-random ghost behaviour.
The Ms Pac-Man vs Ghosts League is an AI competition based on the game. The organisers have released a framework which emulates Ms Pac-Man and allows users to write agents in Java.
Monte Carlo tree search is a decision making process which uses a heuristically-guided sampling method to cut down the search space.
This project is a game-playing agent for the Ms Pac-Man vs Ghosts League framework, using Monte Carlo tree search as its main decision-making process. It was written by Stewart MacKenzie-Leigh as part of a summer research placement in the Computer and Information Sciences department of the University of Strathclyde. The code is based loosely on initial work by Matthias Lenz, and the research was carried out with Dr John Levine. We wrote a paper for PlanSIG 2012, also available here.
If you want to read the source code to see what it does, src/pacman/entries/pacman/MyPacMan.java is a good place to start.
The agent source code is released under the MIT License; you can find a copy in the LICENSE file. Please note that the framework is included as the PacManVsGhosts6.2.jar file, and has a separate license available inside the jar in copyright.txt. The data/ directory is also part of the framework and shares the same license.
You can do what you like with the code - feel free to drop me a line to tell me what you're doing with it.