Coevolutionary automatic game content generation of physics and flighting style games

程序内容生成（PCG）是一种自动生成电脑游戏内容的过程。因此，它具有显著降低游戏开发成本的潜力。商业产品可用于自动生成视觉内容。此外，在过去五年中，PCG在描述游戏中遇到的材料的结构布局方面取得了一些重大进展。例子包括冒险类电脑游戏中自动生成的“迷宫般的世界”，平台类游戏（如《超级马里奥兄弟》）或物理解谜类游戏（如《割绳子》）的特定内容。在这项工作中，我们为两种游戏类型开发了PCG方法：1)《愤怒的小鸟》风格的游戏，其目标是提供尽可能多样化的内容；2)格斗游戏，其内容是由优秀对手的可用性驱动的，这些对手能够抵抗玩家预测其移动的能力。我们对于为《愤怒的小鸟》物理风格游戏定义PCG引擎的见解是，定义解决方案轨迹，并使用它来确定构建内容的约束条件。内容的变化将来自于假设内容生成的随机过程，而约束的使用有助于指导变化算子有效地识别最有希望的内容。飞行游戏的PCG将利用协同进化过程自动识别各种各样的飞行代理策略。将研究参数化约束，以鼓励不同战斗风格的发展。此外，我们也有兴趣扩展战斗游戏中典型的两两模式，这样就可以同时使用几个多对手的战斗模式，从而将这类游戏扩展到两两模式之外。

Procedural content generation (PCG) represents an automatic process by which content for computer games is automatically created. As such, it has the potential to significantly reduce the cost of game development. Commercial products are available for automatically generating visual content. Moreover, there have been several results over the last five years in which significant advances have been made to PCG for describing the structural layout of material encountered in a game. Examples include the automatic generation of 'maze like worlds' as encountered in the adventure genre of computer games, and specific instances of content for platform style games - such as 'Super Mario Bros' - or physics puzzle style games such as 'Cut the Rope'. In this work, we develop PCG approaches for two gaming genres: 1) an Angry Birds style game in which the goal is to provide as diverse a range of content as possible, and 2) fighting games in which content is driven by the availability of good opponents that are able to resist the player's ability to predict their moves. Our insight for defining the PCG engine for the Angry Birds physics style game is to define solution trajectories and use this to identify constraints from which content is constructed. Variation in content will come from assuming a stochastic process for content generation, whereas the use of constraints helps guide variation operators to efficiently identify the most promising content. PCG for flighting games will make use of coevolutionary processes for automatically identifying a diverse range of strategies for flighting agent strategies. Parameterization constraints will be investigated to encourage the development of distinct fighting styles. Moreover, we are also interested in extending the pairwise model of combat typically assumed for fighting games so that several simultaneous multi-opponent modes of combat are available, thus extending the genre beyond pairwise play.