| 结合行为树与Q-learning优化UT2004中agent行为决策 |
| |
| 引用本文： | 刘晓伟,高春鸣. 结合行为树与Q-learning优化UT2004中agent行为决策[J]. 计算机工程与应用, 2016, 52(3): 113-118 |
| |
| 作者姓名： | 刘晓伟 高春鸣 |
| |
| 作者单位： | 1.湖南大学 信息科学与工程学院，长沙 4100122.湖南大学 数字媒体研究所，长沙 410012 |
| |
| 摘 要： | 针对FPS游戏UT2004中的NPC（Non-Player-Character，即非玩家角色）的行为决策不够灵活多变，不够智能等问题，结合行为树与Q-learning强化学习算法，提出了一种预处理与在线学习结合的方式优化NPC行为决策的方法。通过在行为树上的强化学习，NPC行为决策更为灵活、智能，即human-like。实验结果表明了该方法的有效性与可行性。
|
| 关 键 词： | 行为决策 游戏人工智能（AI） Q学习 强化学习 行为树 |
Optimization of agent behavior decision in UT2004 combined with behavior tree and Q-learning algorithm |
| |
| LIU Xiaowei,GAO Chunming. Optimization of agent behavior decision in UT2004 combined with behavior tree and Q-learning algorithm[J]. Computer Engineering and Applications, 2016, 52(3): 113-118 |
| |
| Authors: | LIU Xiaowei GAO Chunming |
| |
| Affiliation: | 1.College of Information Science and Engineering, Hunan University, Changsha 410012, China2.Institute of Digital Media, Hunan University, Changsha 410012, China |
| |
| Abstract: | In FPS game UT2004, NPC’s（Non-Player-Character） behavior decision is not flexible and not smart. Combined with behavior tree and Q-learning algorithm, an improved intelligent behavior decision mechanism is proposed to refine NPC’s behavior decision in way of combination of offline and online learning. Through reinforcement learning on the behavior tree, NPC’s decision becomes more and more smart and flexible, i.e. more human-like. The experimental results show that the method is feasible and efficacious. |
| |
| Keywords: | behavior decision game Artificial Intelligence（AI） Q-learning reinforcement learning behavior trees |
| 本文献已被 万方数据 等数据库收录！ |
| 点击此处可从《计算机工程与应用》浏览原始摘要信息 |
|
点击此处可从《计算机工程与应用》下载全文 |
