Companion players for immersive computer games: How to learn behavioural responses from visual effects

沉浸式电脑游戏的同伴玩家：如何从视觉效果中学习行为反应

• 批准号: 2107651
• 金额: --
• 依托单位: Royal Holloway University of London
• 依托单位国家: 英国
• 项目类别: Studentship
• 财政年份: 2018
• 资助国家: 英国
• 起止时间: 2018 至 无数据
• 项目状态: 已结题
• 来源: https://gtr.ukri.org/projects?ref=studentship-2107651
• 关键词: Companion; players; immersive; computer; games

The proposed research will explore the relevance of AI technological innovation to develop adaptivecompanions for immersive computer games. Given the visual aspects of a game such as characters, objects,surfaces and texture, companions will be developed automatically by tackling hard challenges in gaming suchas Reinforcement learning or other AI and machine/deep learning techniques. The outcomes of the resultingexploration will be fed back to the design process in the form of visual interactions and learned behaviouralresponses to improve the overall immersive experience of the user(s).In particular, the project will address how a computer program can learn from people, both by intelligentimitation and by learning intelligent responses or companion actions. A promising recent line of research is thediscovery of a mathematical correspondence between Generative Adversarial Networks (GANs) - a recentlydeveloped approach to learning to generate new examples of data similar to those in an existing corpus - andInverse Reinforcement Learning (IRL), a technique for learning to imitate the behaviour of a 'teacher' byinferring the teacher's goals and costs from its behaviour. This line of work will allow artists and programmersto find new ways on how agents interact with the environment/world they are creating and even automatesome of the animation of characters that takes a lot of time to develop.

这项拟议的研究将探索人工智能技术创新与开发沉浸式电脑游戏的适应性同伴的相关性。给定游戏的视觉方面，如角色、物体、表面和纹理，同伴将通过解决游戏中的困难挑战而自动开发，如强化学习或其他人工智能和机器/深度学习技术。探索的结果将以视觉交互和习得的行为反应的形式反馈到设计过程中，以改善用户的整体沉浸式体验(S)。特别是，该项目将解决计算机程序如何通过智能模仿和学习智能反应或伴随动作向人学习。最近一个很有希望的研究方向是发现了生成性对抗网络(GANS)和反向强化学习(IRL)之间的数学对应关系。GANS是最近开发的一种学习生成类似于现有语料库中的数据的新样本的方法，IRL是一种通过从教师的行为推断教师的目标和成本来学习模仿教师行为的技术。这一系列的工作将允许艺术家和程序员找到新的方法，了解代理人如何与他们正在创建的环境/世界互动，甚至可以自动完成需要大量时间开发的角色动画。