Application of the Monte-Carlo Tree Search to Multi-Action Turn-Based Games with Hidden Information
Abstract
Traditional search algorithms struggle when applied to complex multi-action turn-based games. The introduction of hidden information further increases domain complexity. The Monte-Carlo Tree Search (MCTS) algorithm has previously been applied to multi-action turn-based games, but not multi-action turn-based games with hidden information. This thesis compares several Monte Carlo Tree Search (MCTS) extensions (Determinized/Perfect Information Monte Carlo, Multi-Observer Information Set MCTS, and Belief State MCTS) in TUBSTAP, an open-source multi-action turn-based game, modified to include hidden information via fog-of-war.
Document Details
- Document Type
- Technical Report
- Publication Date
- Mar 25, 2021
- Accession Number
- AD1134590
Entities
People
- Connor M. Pipan
Organizations
- Air Force Institute of Technology