Improved UCT Algorithm for WTN-EinStein Würfelt Nicht! Game Combining Domain Knowledge

EinStein würfelt nicht! (EWN) game attracts the participation of the majority of scholars relying on randomness and strategy. From the perspective of the global optimization, the UCT algorithm and the domain knowledge are introduced to apply in EinStein würfelt nicht! game agent. First, the evaluation factor C
cur
and the sub-optimal node balance factor C
bal
are introduced to maintain the balance of depth search and breadth search. Second, the domain knowledge for simulation and establishing the tree based on probabilistic rules and natural characteristics of EWN is presented. Finally, a complete EinStein würfelt nicht! game agent (AKe) is constructed. AKe contrasted with other five algorithm agents (Random, SA, SE, UCT1, UCT2), which proved the effectiveness of the improvement. Ake has won the champion of the 2021 Chinese University Student Computer Games Competition with an 87% winning rate, which further validates the significance of the algorithm.