Abstract: In design of modern computer games, it’s very important to model a virtual human with believable perception behaviors. In previous researches, the range of a perception system is always unchangeable. An adaptive perception system based on Q-Learning is proposed, a virtual human confirms the range of a perception system dynamically according to his appraises to the objects in the environment. A demo system of a virtual human searching for specific herbs is realized on PC. The results show that the model can drive a virtual human to have believable perception behaviors.

Key words: Virtual human, Perception, behavior animation, Q-learning.

摘要： 在现代计算机游戏的设计中，建立具有感知行为可信的虚拟人是至关重要的。以往的研究中，虚拟人的感知范围往往是固定的。提出了一种基于Q学习的虚拟人自适应感知模型，虚拟人可以通过对环境中感知对象的评价来动态确定感知范围，并在微机上实现了一个虚拟人找药的原型系统。实验结果表明，该模型能使虚拟人的感知行为具有可信性。

关键词: 虚拟人, 感知, 行为动画, Q学习