상세정보
학술대회
Reinforcement Learning of Intelligent Characters in Fighting Action Games
- 발행일
- 200609
- 출처
- International Conference on Entertainment Computing (ICEC) 2006 (LNCS 4161), v.4161, pp.310-313
- 협약과제
- 06MC1700, 멀티코아 CPU 및 MPU기반 크롯플랫폼 게임기술 개발, 양광호
- 초록
- In this paper, we investigate reinforcement learning (RL) of intelligent characters, based on neural network technology, for fighting action games. RL can be either on-policy or off-policy. We apply both schemes to tabula rasa learning and adaptation. The experimental results show that (1) in tabula rasa leaning, off-policy RL outperforms on-policy RL, but (2) in adaptation, on-policy RL outperforms off-policy RL. © IFIP International Federation for Information Processing 2006.
- KSP 제안 키워드
- Action Games, International federation, Learning and adaptation, Network technology, Reinforcement Learning(RL), information processing, neural network, off-policy