As part of my capstone project, my team explored reinforcement learning (RL) to teach an agent to play the board game Tigers and Goats (HuliGutta). This game’s asymmetric rules and strategic depth made it a good testbed for developing adaptive strategies. We focused on implementing and comparing different RL algorithms, such as value iteration and a 2-step look-ahead approach, to help the agent evaluate board states and plan moves effectively.
I contributed to implementing and testing the game environment and state representation in Python, as well as coding the logic for valid moves, reward assignment, and game outcomes. I also helped run training experiments using the OpenSpiel library and the UH Koa high-performance computing cluster, which allowed us to train and analyze our agents on a larger scale. My work helped ensure the agent could learn from feedback and improve its play over time.
This project strengthened my skills in Python programming, reinforcement learning concepts, and working on a larger collaborative research codebase. It also gave me experience using advanced tools like OpenSpiel and high-performance computing, which taught me how to manage computational experiments and analyze their results.