Monte-Carlo Tree Search in TOTAL WAR: ROME II's Campaign AI

Abstract

This article describes the implementation of Monte Carlo Tree Search (MCTS) in the campaign AI of the game TOTAL WAR: ROME II. MCTS is a search algorithm that combines elements of breadth-first search and random sampling. It has proven successful in board games like Go, and its implementation in TOTAL WAR: ROME II represents its first foray into the AAA games industry. The authors discuss the design motivations for using MCTS, the architecture of the campaign AI, the performance optimizations employed, and the use of domain knowledge to improve MCTS’s effectiveness. They describe how MCTS is used within the AI to make decisions about resource allocation and action selection. The authors also highlight the challenges of implementing MCTS in a complex game environment and the strategies used to overcome these challenges, such as aggressive pruning, eliminating duplication, and soft restrictions. Finally, they emphasize the importance of domain knowledge in guiding MCTS and improving its performance. – AI-generated abstract.