In West Harlem, New York City, Columbia University researchers are developing the Urban Hybrid Twin to study how real-time sensing and traffic models can support safer, more efficient urban transport. The work is part of the three-year project Hybrid Twins for Urban Transportation: From Intersections to Citywide Management, launched in 2021. It focuses on intersections and, ultimately, traffic management across larger areas. 1114
The project combines data from cameras, LiDAR and wireless sensing with traffic and communications simulations. Its testbed is COSMOS, a city-scale wireless research platform deployed in West Harlem. The researchers integrate traffic, network and autonomous-driving simulators, including SUMO, NS-3 and CARLA, with machine-learning methods. The aim is to create a model that uses both live observations and established traffic knowledge to represent and predict conditions on real streets. 1114
The team is investigating applications such as forecasting traffic states, coordinating traffic signals and warning vulnerable road users about potential hazards. These are intended outcomes, not evidence that the system has already delivered citywide improvements. The project’s research has included trajectory-prediction models trained on real-world tracking data from COSMOS, and the team presented AI-based urban traffic work at CVPR 2024. 1213
Columbia’s Xuan Sharon Di leads the research, working with Qiang Du, Zoran Kostic and Gil Zussman; Columbia’s Data Science Institute is also involved. The project’s broader goal is to test whether a hybrid of data-driven learning and traffic modelling can help managers assess conditions and evaluate interventions before applying them on the street. 14