적응형 NPC를 생성하는 행동 정보 관리 기법

다국어 초록 (Multilingual Abstract)

Although people have had more opportunities to enjoy various types of game, most of players have felt less satisfaction with the games. This is the reason that since most Non-player Characters (NPCs) are simple and uniform, they have some limitations for competing with a variety of players. Thus, technologies for creating intelligent NPCs that can compete with each player at a similar level are required. In this paper, we present an action information management method for creating adaptive NPCs based on the algorithm for calculating their action efficiency. This algorithm is useful to the adaptation method for saving and controlling player-appropriate action. In our method, adaptive NPCs observe the actions of players and collect the relationship data between status and action. The efficiency value of the action data is calculated and data of similar status are grouped, and finally stored at the action database. The game system of NPC updates the action database and stores diverse actions. Then, NPC selects action with high efficiency value. We have tested our algorithm on an action game. A random test subject performed a one-on-one game against an adaptive NPC in real-time. As a result, the action dispositions of both the subject and NPC are analyzed in a log file to determine whether or not the disposition of the subject is similar to that of the NPC. The statistics of the diverse test results shows that NPCs become adaptive to players with error rate within less than 6%.