| Summary: | 碩士 === 國立高雄大學 === 資訊管理學系碩士班 === 98 === Game is one of the indispensable activities for humanity lives. In recent years, the development of technology also brought huge market to game industry. One of the appealing reasons in game is that the player can interact with the non-player-characters in game. Artificial intelligence is very important for these non-player-characters due to Artificial intelligence can let non-player-characters have more interactions with players.
The environment in digital games is changing continuously, so it is a challenge to add artificial intelligence in non-player-characters. In this research, we would like to use reinforcement learning in non-player-characters’ artificial intelligence. Reinforcement learning is a un-supervise learning method, and it is usually used in automatic machine learning process. Reinforcement learning is a trial-and-error process, and the agent will change his actions by exploring the new environment.
The most difficult to apply Reinforcement learning in digital games is that the method requires a long learning time . In this research, we use fuzzy theory to increase the learning efficiency. The results of this research experiment also prove the improvement of learning efficiency through using fuzzy reward to replace fixed reward. Different types of game need different settings of reward. In practice, the probability to apply reinforcement learning in digital games can be enhanced once the suitable reward mechanism has been found out.
|
|---|
