Skull and Roses: Multi-agent Reinforcement Learning approach of modelling Bluffing and Theory of Mind

Ditu, Alexandru Dumitru (2024) Skull and Roses: Multi-agent Reinforcement Learning approach of modelling Bluffing and Theory of Mind. Bachelor's Thesis, Artificial Intelligence.

Preview	Text Thesiss4004027DituAlexandruMulti-AgentReinforcementLearningApproachOfModellingBluffingandTheoryOfMind-1.pdf Download (1MB) | Preview
	Text Toestemming Ditu.pdf Restricted to Registered users only Download (155kB)

Abstract

This study explores the possible benefits of using a bluffing strategy in the bidding phase of the game Skull and Roses, an imperfect information game that has many similarities with other games such as Poker. By simulating various game scenarios using Python and reinforcement learning agents, the research aims to determine whether bluffing provides a statistical advantage to the player when compared to strategies that do not employ bluffing. Essential concepts such as game theory, reinforcement learning, Q-Learning and Theory of Mind (ToM) are discussed in order to provide a theoretical foundation for the study. I hypothesize that bluffing will outperform the non-bluffing strategies or at the very least provide similar performance to them. The results were extracted over a total of 100 individual simulations, containing 1000 game rounds each, for each game scenario. After analysis, the results show that the agents that are using the bluffing strategy are slightly outperforming those that do not, demonstrating a statistically significant advantage. Therefore, this outcome is in line with the hypothesis. However, a key discovery has been made in the testing phase: bluffing is only beneficial when used alongside specific card layouts such as the last placed card being a Skull. This hints that there is a very close relation between bluffing and the first phase of the game. These findings contribute to a deeper understanding of strategic decision-making in complex, imperfect info

Item Type:	Thesis (Bachelor's Thesis)
Supervisor name:	Weerd, H.A. de
Degree programme:	Artificial Intelligence
Thesis type:	Bachelor's Thesis
Language:	English
Date Deposited:	09 Aug 2024 09:09
Last Modified:	09 Aug 2024 09:09
URI:	https://fse.studenttheses.ub.rug.nl/id/eprint/33905

Actions (login required)

View Item