为构建数字化虚拟世界，需要解决虚拟场景高效率、高质量构建的难题，从而推动虚拟现实技术在重要行业领域中的应用。针对手动构建效率低、成本高，而现有自动构建技术结果可控度低、难以反应需求意图、关键细节不精准、动态生成不自然等问题，本项目提出以自然交互为手段，将使用者（人类）的知识和意图融入智能计算过程，实现“知识+计算”优势互补，构建“自然交互驱动的虚拟场景内容智能构建”理论系统。针对项目关键科学技术问题，重点开展自然眼动交互及其牵引的多模交互分析、交互意图理解引导的虚拟场景图像生成、交互行为学习驱动的虚拟人类对象构建等方面的基础研究，力争取得相关算法与方法创新，初步形成眼动、言语等多模态联合交互驱动的虚拟场景图像和人类对象智能构建理论系统，为典型应用提供关键技术，推动虚拟现实技术向智能化方向发展。

In order to build up the virtual world, the requirements on high-efficiency and high-quality virtual scene construction should be fulfilled, and this is important for the VR technology being used in many important applicational fields. Conventional manual methods are slow and costly, while existing automatic methods face difficulties such as low controllability, insufficient expressiveness, inaccurate details and unnatural behaviors. We propose a new solution based on natural interaction, which combines the human knowledge and intention into the intelligent computing process and makes them benefit each other. This leads to the theoretical solution of natural interaction-driven intelligent construction of virtual scenes. In order to solve its key theoretical and technical problems, we conduct researches including gaze tracking-driven natural human-computer interaction, human intention-guided virtual image synthesis, human action learning-based dynamic generation of virtual person. With the expected achievements, we will deliver new systems that generate virtual scene images and dynamic virtual humans based on eye/speech/gesture-interactions, and they are expected to be helpful for many important applications and also help promote the development of VR+AI.