This Java-application contains all required components to simulate a game of Ms. Pac-Man and let an agent learn intelligent playing behaviour using reinforcement learning and either Q-Learning or SARSA.
The framework was developed by Luuk Bom and Ruud Henken, under supervision of Marco Wiering, Department of Artificial Intelligence, University of Groningen. It formed the basis of a bachelor's thesis titled "Using reinforcement learning with relative input to train Ms. Pac-Man", L.A.M. Bom (2012).
- (Graphical) simulation of the game Ms. Pac-Man
- Uses a small number of clever relative inputs to represent the game environment
- Reinforcement learning with either Q-Learning or SARSA
- Highly configurable and easy to tweak
- Results can be exported to .CSV or plotted in graphs
- Extensive documentation
Be the first to post a review of Ms. Pac-Man Framework!