Interactive Storytelling and Real-time Heuristic Search

1.如果有机会与观众互动，人类讲故事的人是适应性强的。睡前故事和个性化辅导是根据观众调整故事内容以最大限度地发挥其有效性的例子。这样的精品交付对于传统的大众市场是不可扩展的，因此，商业电影和大型讲座遵循一刀切的原则。计算机辅助媒体，如视频游戏，可以结合两个世界的最好，大规模地为观众的每个成员提供定制的娱乐。此外，观众可以成为故事的积极参与者，通过自己的行动与叙事互动。最近的研究从两个角度探讨了互动讲故事的问题。首先，叙事以一种正式的语言表示，因此自动策划人员可以适应观众的行动，同时尊重作者的叙事目标。其次，自动获取的观众偏好模型被用来为他们量身定做住宿。我的研究小组是第一批在交互式数字故事讲述中使用自动观众建模的小组之一。在拟议的研究计划下，我们将首先扩展我们现有的系统，建立一个观众情绪反应的明确模型。这将允许我们的自动讲故事者明确地定制其叙事，以引起观众一系列特定的情感反应(例如，喜悦紧随其后的希望)。其次，我们将使用我们的体验经理的能力来自动生成大量故事情节，用于叙事空间探索。我们将从构建一个自动的故事设计师助手开始，以预先估计观众对互动故事空间的反应。然后，我们将添加一个自动修改故事空间的机制，以确保期望的观众反应。这项研究的好处包括更具沉浸感的个性化娱乐和培训系统，更可信的虚拟角色和更低的开发成本。2.人工智能代理经常需要在有限的时间内做出信息不完全的决策。应用范围从路线规划到个人数字助理，再到视频游戏中的不可玩角色。实时启发式搜索是一种在不确定条件下进行实时决策的方法，具有很强的数据局部性，特别适合于并行性。到目前为止，我的研究小组已经产生了一系列最先进的实时启发式搜索算法。我们通过将几个关键范例引入实时启发式搜索领域来实现这样的性能。随着计算硬件变得无处不在的并行，我的研究小组将把我们的数据库驱动的实时启发式搜索扩展到并行硬件。我们将对产生的算法进行实证分析，并继续我们在实时启发式搜索理论方面的工作。这项研究的好处包括更智能的自主代理，特别是互动娱乐中更可信的不可玩角色。

1. Given an opportunity to interact with their audience, human storytellers are adaptive. Bedtime stories and individualized tutoring are examples of adapting story content to the audience in order to maximize its effectiveness. Such boutique delivery is not scalable for traditional mass markets and, as a result, commercial movies and large-scale lectures follow the one-size-fits-all principle. A computer-assisted medium such as video games can combine the best of both worlds, delivering entertainment customized to each member of the audience on a mass scale. Furthermore, the audience can become an active participant in the story, interacting with the narrative through its own actions. Recent research has attacked the problem of interactive storytelling from two angles. First, narratives have been represented in a formal language so that automated planners can accommodate the audience's actions while respecting the author's narrative goals. Second, automatically acquired models of the audience's preferences are used to tailor the accommodations to them. My research group was one of the first to use automated audience modeling in interactive digital storytelling. Under the proposed research program we will first extend our existing systems with an explicit model of the audience's emotional response. This will allow our automated storyteller to customize its narrative expressly to elicit a series of specific emotional responses from the audience (e.g., hope followed by joy). Second, we will use the ability of our experience managers to automatically generate a multitude of story plots for narrative space exploration. We will start by building an automated assistant to story designers to estimate audience's response to the interactive story space beforehand. We will then add a mechanism to automatically modify the story space to ensure desired audience response. The benefits of this research include more immersive personalized entertainment and training systems, more believable virtual characters and lower development costs. 2. Artificial Intelligence agents frequently need to make decisions with incomplete information in limited time. The applications range from route planning to personal digital assistants to non-playable characters in video games. Real-time heuristic search is an approach to real-time decision-making under uncertainty and is particularly amenable to parallelism due to its strong data locality. To date, my research group has produced a progression of state-of-the-art real-time heuristic search algorithms. We achieved such performance by introducing several key paradigms to the field of real-time heuristic search. As computing hardware has become ubiquitously parallel, my research group will extend our database-driven real-time heuristic search to parallel hardware. We will analyze the resulting algorithms empirically as well as continue our work on the theory of real-time heuristic search. The benefits of this research include more intelligent autonomous agents and, in particular, more believable non-playable characters in interactive entertainment.