What happens when artificial intelligence starts controlling the flow of a city?

→

Traffic simulations are projected onto a physical model of Lucerne. In the simulation, models trained with reinforcement learning control the traffic lights. Visitors can trigger scenarios such as accidents or event traffic and observe how these situations affect traffic flow, waiting times and the behaviour of the system.

Before the AI can decide, the city has to become a simulation.

→

SUMO simulates the movement of vehicles and scenario-based disruptions, while reinforcement learning controls the traffic lights. Simulation data is sent via Python/TraCI, translated in TouchDesigner and projected back onto the physical urban model.

→

Using SUMO-RL and Ray RLlib, the traffic light systems were trained with reinforcement learning. Each agent observes traffic density, waiting vehicles and signal phases, then decides whether to keep the current phase or switch to the next one.

→

A selected section of Lucerne was translated into a physical city model using QGIS, Illustrator, laser-cut MDF plates and plexiglass elements. Fibre optic cables were integrated into the traffic lights to visualise their live signal states, while the model acts as a neutral projection surface where traffic flows, events and signal states become visible.

→

Through a physical control panel, visitors can choose different traffic scenarios, from everyday traffic to accidents or Fasnacht. Each scenario changes the behaviour of the simulation and triggers visual, animated and audio-based explanations.

→

The visual language is inspired by architectural models, Swiss Design, traffic signage and reduced cartography. Colours, glow effects and typography are used functionally to highlight relevant events, routes and system states without overloading the installation.

→

The project combines traffic simulation, reinforcement learning, projection mapping, animation, sound, microcontroller interaction and physical model-making. Instead of presenting AI as an invisible technical solution, it turns algorithmic decision-making into something spatial, observable and debatable.