@techreport{oai:ipsj.ixsq.nii.ac.jp:00218737,
 author = {奥田, 真 and 池田, 心 and 橋本, 剛 and Makoto, Okuda and Kokolo, Ikeda and Tsuyoshi, Hashimoto},
 issue = {7},
 month = {Jun},
 note = {ターン制戦略ゲームの AI は TUBSTAP というプラットフォームで研究が行われているが，AI の強さは初心者プレイヤと互角程度に留まっている．その理由の 1 つとして，TUBSTAP の複数着手性が挙げられる．1 手で複数の駒を動かすルール上，合法手が多すぎて深く読めなくなるのである．このため，TUBSTAP ではモンテカルロ木探索が有望とされ，枝刈りなどノードを減らす工夫を行う研究がされてきた．一方でノードを減らす以外の工夫としてプレイアウトの方策改良があるが，TUBSTAP でこれを検討した研究は少ない．TUBSTAP における従来のプレイアウト方策には，駒の移動行動と攻撃行動を事前に決めた確率で選択するものがあるが，本論文ではこの確率を攻撃行動を優先するように変えたプレイアウト方策を提案する．対戦実験の結果，提案した方策は従来の方策に対しあるマップで勝率 87.7% で大幅に勝ち越し，提案方策は有効であることが分かった．また，複数マップでの対戦実験により提案方策の影響を調査した．, AI for turn-based strategy games has been studied on the TUBSTAP platform. However, the strength of the AI has remained at about the same level as that of novice players. One of the reasons for this is the multiple-move nature of TUBSTAP. Due to the rule of moving multiple pieces in one move, there are too many legal moves to read deeply. For this reason, Monte Carlo tree search is considered promising for TUBSTAP, and research has been conducted to reduce the number of nodes by pruning branches and so on. On the other hand, there is another way to reduce the number of nodes, which is to improve the playout strategy, but few studies have examined this in TUBSTAP. The conventional playout strategy in TUBSTAP is to select a piece's move or attack action with a predetermined probability, but in this paper, we propose a new strategy to select a piece's move or attack action with a predetermined probability. In this paper, we propose a playout strategy that changes this probability to give priority to attacking actions. In a competitive experiment, the proposed strategy significantly outperformed the conventional strategy on one map by 87.7 percent, and the proposed strategy was found to be effective. We also investigated the impact of the proposed strategy in a competitive experiment on several maps.},
 title = {ターン制戦略ゲームにおける攻撃優先プレイアウトの影響},
 year = {2022}
}