Authorial Control of Monte Carlo Tree Search Agents in Games

随着人工智能(AI)的发展，数字游戏代理的开发越来越受到学术界的关注。由于策略博弈在博弈状态空间和动作空间方面的复杂性，智能智能体的实现一直被认为是一个具有挑战性的课题，现有的智能体设计通常采用蒙特卡罗树搜索(MCTS)等搜索算法。然而，复杂程度高等实际问题阻碍了它的应用，也带来了一些开放性的研究问题。拟议的研究将侧重于改进管理和协调制度。我们研究了如何确定MCTS的收敛，如何度量代理的决策，以及如何建立MCTS的意识来实现作者控制。本研究成功完成后，将对学术界和游戏产业产生重大影响，推动进一步的研究方向。

Developing agents for digital games has been receiving increasing attention from academia with the advance of Artificial Intelligence (AI). As the strategy game is complex in terms of the game state space and action space, implementing intelligent agents has been regarded as a challenging topic.Search algorithms, e.g. Monte Carlo Tree Search (MCTS), are commonly used in existing works to design agents. However, practical issues, such as the high complexity, hinder its application and introduce open research problems. The proposed study will focus on improving MCTS. We investigate how to determine the convergence of the MCTS, how to measure agents' decisions and how to build its awareness to realize authorial control. On successful completion, the research will have a great impact on both the academia and game industry, motivating further research directions.