Can’t Stop, Won’t Stop: An MCTS Agent for Two-Player Can’t Stop

In this senior project, I first implement the Can’t Stop game and then explore the intricacies of designing a Monte Carlo Tree Search (MCTS) player tailored specifically for the board game Can’t Stop. After Professor Glenn suggested this project idea and researching the game, I found myself intrigued at the prospect of attempting to create an agent that could capitalize upon information about the game states, simulate different outcomes of different moves, and ultimately make decisions that would allow it to balance the risk and reward of pressing its luck. For this MCTS player, I focused primarily on effectively balancing the importance of opponents’ positions, progress of neutral markers, column length and column probability to determine two aspects of the game: 1. When to stop rolling the die. 2. Which moves to choose when the die were rolled. Following the development of this agent, I tested its capabilities against a baseline Can’t Stop heuristic known as the “Rule of 28.” This strategy evaluates players’ neutral markers progress and roll difficulties and then scores these factors. It then advises them to halt their progression near a score of 28, maximizing the chances of retaining gained ground while minimizing the risk of losing it. Incorporating insights from this heuristic, the goal was to employ techniques learned from CPSC 474 to design and implement an MCTS agent capable of optimizing gameplay strategies, with particular focus on its performance compared to the Rule of 28 Player.