Date of Award
2019
Document Type
Thesis
Degree Name
Bachelors
Department
Natural Sciences
First Advisor
Doucette, John
Area of Concentration
Computer Science
Abstract
This thesis explores the topic of applying reinforcement learning methods to train an autonomous agent to play the video game Super Bomberman. Video games are highly compatible with techniques for reinforcement learning due to the nature of their design. The approach used in this thesis trains an agent using deep Q-learning, a method which has been shown to be effective in related works on using reinforcement learning on video games. The implementation of deep Q-learning onto a Super Bomberman-playing agent was made possible by the use of video game emulation software with advanced control capabilities. By using this advanced emulator, the agent could directly play the original game while using queried values from the emulated console’s memory to observe the game state. The creation of a functional deep Q-learning agent was achievable using this method, but due to high runtime costs this approach is not yet practically useful for training purposes.
Recommended Citation
Gray, Austin, "SUPER BOMBERMANN: Q-LEARNING WITH DEEP ARTIFICIAL NEURAL NETWORKS FOR THE BOMBERMAN VIDEO GAME" (2019). Theses & ETDs. 5697.
https://digitalcommons.ncf.edu/theses_etds/5697