CARLA-Twin: A Large-Scale Digital Twin Platform for Advanced Networking Research

Digital Twins (DT) create virtual replicas of physical entities, enabling large-scale simulations and advanced data analysis in the twin world (a.k.a. the virtual environment). This capability significantly improves decision-making in the physical world, with applications spanning transportation, manufacturing, training, healthcare, etc. The success of DT hinges on fast and reliable networking technologies, which ensure efficient data collection from the physical world and seamless execution of decisions back into it. However, academic research in DT networking often faces challenges due to the lack of access to large-scale physical facilities, such as cities or warehouses. To address this gap, we develop a DT framework–CARLA-Twin–to facilitate networking research, leveraging the popular CARLA simulator–a large-scale autonomous driving research platform. In particular, CARLA-Twin employs a dual-computer architecture, where one instance of CARLA represents the physical world and the other acts as its twin world. These instances are interconnected through a bidirectional communication module to facilitate interactions between the physical and twin worlds. To validate the effectiveness of CARLA-Twin, we conduct a case study of DT-assisted city traffic management. Performance is evaluated under three scenarios: (1) Without DT Assistance, (2) DT with Ideal Communication, and (3) DT with Defective Communication. Simulation results demonstrate that DT scenarios have better performance compared to non-DT ones. Specifically, the collision rate is reduced by 95% under ideal communication scenario, while the average vehicle speed is nearly 2.8 times faster. However, elevated packet loss rates and network latency may diminish the effectiveness of DT systems, showing the importance of advanced networks. We will open-source our implementations.