Unjam: Closed-Loop Traffic Control with Digital Twins

Abstract

This talk traces the technical and organizational journey behind Unjam, a Swiss project developing the Unjam Traffic Gym — a control-oriented digital twin of Zurich’s traffic network built on closed-loop SUMO microsimulations. Beyond the control layer, we share lessons and anecdotes from applying traffic control in the wild: pitching ideas and prototyping within the TalentKick accelerator, securing innovation-oriented funding, setting up legal and technical infrastructure (e.g., IP management, entity formation, NDA coordination), and building custom data pipelines to fetch and synchronize heterogeneous traffic signals — all in close collaboration with the Zurich startup ecosystem.We reflect on how control theory informs both system design and broader architectural choices, offering robustness, modularity, and insight into complex urban dynamics. The talk culminates in a data-driven control framework that combines model-free optimization with interpretable, structure-aware policies, designed for real-world deployment and demonstrated within the Unjam Traffic Gym. Finally, we outline key open challenges, including online state estimation, distributed coordination, integration with legacy infrastructure, and scalability in large urban networks.