Please use this identifier to cite or link to this item: http://hdl.handle.net/10609/119226
Title: Aplicació de tècniques d'aprenentatge computacional per la creació d'agents jugadors de Sushi Go
Author: Montufo Rosal, Jose
Tutor: Nuñez Do Rio, Joan Manuel
Others: Ventura, Carles  
Abstract: The application of reinforcement learning techniques to board games has been the subject in recent years of many projects among the specialized scientific community. The mechanics and rules of board games tend to form an ideal environment to be used as a test bed for the tools provided by the area of reinforcement learning. This project was born in order to use the Sushi Go card game as a basis for creating various intelligent agents capable of learning a strategy that would allow them to be competitive against a human. The goals of the project are to compare performance provided by various reinforcement learning techniques, study the optimal strategy they use, and create a UI that allows users to confront agents. To achieve this goal, a pre-existing implementation has been modified to build a standard OpenAI Gym environment for Sushi Go. Subsequently, the environment has been used to apply the different learning algorithms in the creation of the agents. Finally, the comparison between the agents to determine the most optimal algorithms was performed, and the strategy followed by the best performing agents was described. At the end of the project, the author has challenged in a series of games to the best agent, being able to win almost all. This fact only indicates that agents still have much room for improvement, either by applying new algorithms, or by expanding the state space they use to obtain information from the environment.
Keywords: reinforcement learning
board games
OpenAI
Document type: info:eu-repo/semantics/bachelorThesis
Issue Date: Jun-2020
Publication license: http://creativecommons.org/licenses/by-nc-nd/3.0/es/  
Appears in Collections:Bachelor thesis, research projects, etc.

Files in This Item:
File Description SizeFormat 
jmontufoTFG0620memòria.pdfMemòria del TFG2,22 MBAdobe PDFThumbnail
View/Open
jmontufoTFG0620presentació.pdfPresentació del TFG537,51 kBAdobe PDFThumbnail
View/Open