Incorporating Domain Knowledge Into Monte Carlo Tree Search in Dark Chess

This paper proposes a Dark Chess computer program based on Monte Carlo Tree Search (MCTS). MCTS is a powerful algorithm used in computer games. We introduce a method that utilizes domain knowledge about piece values and positioning to score new nodes in the search tree. This addresses the limitation of MCTS, which performs poorly when nodes have only been visited a few times. Ultimately, this approach improves the strength of the Dark Chess computer program.