1R-05
Applying reinforcement learning to deck building games
○YICHEN CUI,吉浦紀晃(埼玉大)
A reinforcement learning study centered on the modern card game Slay the Spire. This research involves modeling the game environment of Slay the Spire and utilizing neural networks to extract and process its features, applying the Deep Q-learning algorithm to train AI agents. The study analyzes the modeling
requirements of this type of game from multiple perspectives and experiments with different reinforcement learning environment designs to achieve optimal training outcomes. It seeks to explore whether reinforcement learning methods can be effectively applied to this innovative card game genre and further investigates whether AI can provide meaningful assistance in actual game development.