This project aims to build a demo system that shows what future traffic may look like in an urban digital twin world. For traffic systems, future conditions can already be estimated by deep learning models. However, these models usually produce summarized results, such as average traffic conditions over the next few minutes or colored congestion levels on a map (e.g., red refers to heavy congestion). Although these outputs are useful for analysis, they do not match the kind of digital twin experience we expect, where people can directly see how vehicles may move and interact in a visible future world.
To address this limitation, researchers have started to explore macroscopic traffic simulation algorithm based on future vehicle route information [1,2]. Compared with summarized prediction results, these methods can provide more detailed and interpretable future traffic states. For example, they can estimate how many vehicles will be on a road at a specific future time, how many may be waiting near an intersection, and how traffic may be distributed across a road network. However, these results are still mainly numerical or structural. They are closer to future traffic itself, but still do not directly show what the future traffic scene may actually look like in a digital twin system.
Inspired by recent AI video generation tools [3,4], they have shown strong ability in creating realistic road scenes and videos from structured inputs, such as road map, vehicles' positions, and other scene descriptions. However, these tools are mainly designed for visual generation. They are not directly connected to real traffic estimation systems or urban digital twin applications.
Based on this background, the goal of this project is to connect future traffic estimation with visual generation tools and build a complete demo pipeline for urban digital twins. Students will build on an existing traffic simulation framework that estimates future traffic conditions from vehicle future route information, and then design a presentation system that converts these estimated traffic states into inputs that video generation tools can understand. These inputs may include road layouts, approximate vehicle locations, and simple vehicle boxes, which can then be used by rendering tools or existing AI video generation tools to produce images or short videos.
Through this project, students will have the chance to learn about traffic simulation, future traffic estimation, scene design, and AI-based visual generation in an accessible and practical way. They will also gain experience in building a full system prototype and presenting it as an interactive demo. If the project goes well, there may also be an opportunity to further develop it into a demo paper for a top conference.