To Create Intelligent Adaptive Neuro-Controller of Game Opponent from UCT-Created Data

Game AI controlled by UCT which achieves excellent performance in computer go can be applied to control non-player characters (NPCs) in video games. While, it is computation intensive algorithm, so applying it to on-line game is not suitable. But data collected from NPC controlled by UCT is able to be utilized to train neuro-controler. Furthermore, neuro-controler is an efficient algorithm due to its capability of extracting knowledge from training data which is generated from UCT. In order to obtain outstanding performance of neuro-controler, training data is a key factor but the structure of neuro-controler is also important. In this paper, the prey and predator game genre of dead- end is utilized as a test-bed, the basic principle of UCT and neuro-controller is drawn, and the effectiveness of their application to game AI development is demonstrated.