Playing the Game of Skull - A Multi-Agent Deep Reinforcement Learning approach

Kapusheva, Maria (2023) Playing the Game of Skull - A Multi-Agent Deep Reinforcement Learning approach. Bachelor's Thesis, Artificial Intelligence.

Preview	Text Bachelor_s_Thesis_Maria_Kapusheva (2).pdf Download (619kB) | Preview
	Text toestemming.pdf Restricted to Registered users only Download (127kB)

Abstract

The Game of Skull is a multi-player board game, which similarly to poker and other bluffing games, is characterized by its partially observable outcomes. While easy to learn for humans, it poses a challenge to Artificial Intelligence algorithms due to the partial observability and its game mechanics - it has a large number of short stages where different actions are legal. In this paper, we are investigating to what extent Deep Q-Network agents could learn how to play the Game of Skull. Furthermore, by adapting and incorporating the scaffolding learning technique from the field of psychology with our Multi-Agent Deep Reinforcement Learning methods, we are researching if these methods are an effective tool to learn the game and how they compare to the Vanilla DQN agents. According to the results outlined in this paper, the agents successfully learn to play the game and consistently reach the final stage, however, using those algorithms results in deterministic agents that deploy rigid strategies, and can hardly adapt to new playing styles. In addition to that, the agents that learned through the scaffolding technique perform slightly better than Vanilla DQN agents, which is a possible direction for future research on the topic.

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:	22 Mar 2023 09:38
Last Modified:	22 Mar 2023 09:38
URI:	https://fse.studenttheses.ub.rug.nl/id/eprint/29461

Actions (login required)

View Item