Q-Learning implementation for GVG-AI game number 80

🏢 University of Huelva (UHU)
📆 course 2020-2021
🎓Artificial Intelligence
Ihar Myshkevich (@IgorMy)

Introduction

Two simple Q-Learning implementations for GVG-AI game number 80. In the first implementation, everything the agent is capable of perceiving is used as a state. This method tends to overfeeding, so this agent only can resolve the map in which it was trained. In the case of the second implementation, the space of states is reduced to 64 states, allowing the agent not only to learn quickly, but also to overcome all the maps without problems.

Video explanation in Spanish

https://www.youtube.com/watch?v=K6iVxWDEhF8&feature=youtu.be

Programs used

Eclipse IDE 2020‑12

How to try it

1. Clone the repository
2. Add the project to Eclipse
3. Build it
4. Run SinglePlayer.Test