【作者】 周强；

【导师】 高春鸣；

【作者基本信息】 湖南大学 ， 计算机应用技术， 2012， 硕士

【摘要】 面对不断成熟的多元化游戏市场，游戏开发人员希望通过游戏智能设计上的改进来提高游戏的可玩性。当前游戏的智能设计中，除了行为的外在表现以及寻路算法的优化外，智能角色的控制也是智能开发人员的研究重点。开发人员希望游戏智能角色能够模拟游戏中玩家行为和策略的方式，从而缩小现实与虚拟间的差距增加游戏的挑战性。本文采用基于模型的智能行为系统，并结合认知科学中的SOAR行为模型以及经济学中的有限理性决策理论，使得游戏中智能角色的感知、决策以及行为执行过程更加贴合玩家的行为模式。本文的主要工作及取得的成果：(1)针对过去视锥视距布尔化感知域的设定方式中的问题，本文首先引入生物视觉感知模型中的感受野理论重构了智能角色的可确定性感知域，然后结合视觉敏感度模型来计算非确定性感知域的确定性值，最终综合分析前面两个部分的信息生成智能角色的感知输入信息。对抗性实验表明，改进后的感知模型能够在不影响系统效率的前提下为后续智能决策过程提供更加充分且有用的信息。(2)本文采用前瞻树递归预测的方法来提高智能角色的决策能力。该方法将在每次系统更新时都会依据智能角色的属性状态转化自动构建一棵前瞻树，同时结合偏好关系以及效用评估理论来计算前瞻树中每一个执行子的效用值大小。最后，系统依据前瞻树顶层执行子的综合效用大小来选择最满意的行为作为输出。分析实验数据发现，改进后的智能决策方法更加贴合玩家的思维方式，并且产生更好的决策效果。(3)在以上智能决策改进方法的基础上，本文采用行为多队列的行为组织方式应对复杂环境下行为执行中的中断、恢复、协作等情况的发生，控制决策行为按照预期的顺序执行。最后，本文通过和当前智能决策方法进行对抗性实验，从胜负关系比和系统效率两个方面来验证改进后智能行为系统的有效性。更多还原

【Abstract】 In the face of the gradually matured and diversified game market, the gamedevelopers expect that the melioration of the intelligence design will enhance thegame’s playability. Apart from the external performance of the behavior and theoptimization of the path-finding algorithm, it is the main concern of game developersthat the intelligent agent is reasonably controlled in the intelligent design. Thedevelopers narrow the gap between virtual and real to improve the game challengingthrough making the intelligent agent simulate players’ behavior and strategies.This paper chooses the model-based intelligent behavior system as the originalsystem, and references the SOAR cognitive science model and the bounded rationalitymodel to make the intelligent agent fit the players’ behavior pattern from3aspects:the perceptron, the decision-making and the behavior control unit. The main work andachievements have been discussed as follows:(1) At first, this paper first reconstructs the intelligent agent’s perception domainthrough introducing in the receptive field of the biological visual perception model, inorder to solve the problem which exits in using cone-horizon boolean to set theperception domain. At the same time, the visual sensitivity model is exploited tocalculate the value of the uncertain perception domain. Therefore, the inputinformation will be determined by comprehensively analyzing the previously data.The confrontational experiments show that the new perceptron can provide a moreadequate and useful information for the decision-making without affecting the systemefficiency.(2) In this paper, the recursive prediction method combining the forward-lookingtree is proposed to improve the decision-making ability. In every time the systemupdate, this method automatically builds a forward-looking tree based on theintelligent agent’s attributes. Meanwhile, the system calculates the utility value ofevery executor in this tree according to the preference relation and the utility function,and then selects the most satisfactory behavior as output through systematic assessingthe comprehensive utilities of the top-level executors. The experimental analysisfound that the improved method fits the players’ thinking patterns, and obtains abetter effect.(3) Based on the above improvement, the behavior multi-queue is constructed to make intelligent agent’s behavior interruptible, resumable and collaborative in thecomplex scenes for behavior planning, so that the decision-making behavior will beexecuted as the expected order.In conclusion, the effectiveness of the improved intelligent behavior system isproved by compering with the current the intelligent behavior systems from thevictory and defeat relations and the system’s efficiency.更多还原