A primary goal of modeling virtual environments for driving simulations is to reproduce the real world as completely as possible. Designers have to accurately represent the geometric information about real roadways and reproduce the real traffic. This study introduced improved methodologies to build large-scale driving simulations filled with autonomous vehicles. The process of modeling the virtual environment is categorized into two levels: geometric and topological. At the geometric level, real-scene based road networks were reproduced in virtual environments with a significantly reduced level of workload through image processing programs with aerial images. At the topological level, this study made innovative explorations of connected driving network with virtual reality as the research approach. Vehicle-to-Vehicle, Vehicle-to-Infrastructure communications were enabled. On the basis of V2X communications, this study modeled 3 standard driving behaviors: car-following, lane-change and intersection, by implementing decision making processes with logic rules formulated as decision trees. Finally, the study provided a demonstration of autonomous driving in Virtual Reality.