| Summary: | 碩士 === 龍華科技大學 === 電子工程研究所 === 99 === In recent years, game artificial intelligence has been making great progress, and the simulation of human-like decision in non-player character (NPC) is also more and more realistic. However, the simulated human emotions in game have been few studies in these days. This is because of today's game simulation of human emotion mostly uses the finite state machine, and the finite number of inputs and states makes it difficult to apply it to simulated the complicated human motions.
In this thesis, we first review and summarize the theories and researches on human emotions and flocking. Then we select the appropriate items from human emotions to implement them in game. We use two popular machine learning methods, neural network and support vector machine, to predict which emotion should be excited for a NPC. The experiment results show that the applications for both neural network and support vector machine achieve the prediction accuracy rate of over 93%. In other words, both of them can successful using in the on-line simulation of emotion in games, and can easily replace the traditional method of finite state machine.
After that, we develop a simulated system which embedded above trained SVM module to simulate the flocking behavior by the infection of human emotions. The experiment results show that when the flock is activated by a special type of emotion, e.g. angry, our simulation system can successfully simulate the non-player characters all have the human-like (or animal-like) behavior.
|
|---|
