Abstract for Monte Carlo Tree Search Agents and Procedural Board Generation in Boreder

This paper explores the development of an AI agent for the self-made game Boreder, a turn-based area-control game set on a procedurally generated 8x8 grid. The AI agent utilizes the Monte Carlo Tree Search (MCTS) algorithm to navigate the game state space and select optimal actions based on simulated playouts. While an attempt was made to incorporate Ford-Fulkerson’s algorithm as a heuristic to determine offensive moves, it led to suboptimal strategies due to overemphasis on blocking paths and lack of long-term planning. This project aims to reveal optimal strategies in Boreder through repeated play across various board configurations, with the overarching goal of developing a game that provides players with a challenging but rewarding play experience. By showcasing the performance of computer agents against human players, this project contributes to the exploration of methodologies for evaluating and refining board game strategy. This project achieves those goals by highlighting the notable trends in play exhibited by the agents. Qualitative observations of the AI agent’s strategies, such as defensive play and mutually destructive moves, are presented. Moreover, this project discusses the implementation of a procedural generation algorithm to randomize map environments for each game. This ensures that the strategies that surface to be independent of the 1 initial state of the board, and are universally applicable across all games of Boreder. Future work includes incorporating additional heuristics to inform the MCTS agent, enabling automated self-play for the collection of game data, and assessing the inherent balance of a map generated by procedural generation algorithm.