Open-loop control of traffic lights in an urban environment is typically tuned so as to generate green waves, which ideally should ensure that vehicles never have to stop, avoiding waste of capacity of the intersections. A green wave minimizes average delay under ideal, noiseless assumptions, by selecting the phase shift between the switching times of successive traffic lights, so that this phase shift corresponds to the travel time between intersections. Achieving green waves is easier if the vehicles travel together in large platoons along major axes of traffic flow. In a large network with dispersed two-way traffic and with different time delays along different roads connecting successive intersections, it becomes difficult to find a perfect green wave switching schedule. Moreover vehicles entering from uncontrolled side streets or parking lots, unexpected delays along connecting link roads, and all other sources of noise also lead to a deterioration of the performance of the open-loop strategy.Local feedback control tries to avoid this deterioration by adjusting the switching times of the traffic lights to currently available data on the expected arrival times of vehicles. This feedback control tends to destroy the green wave synchronization. Combining a green wave with feedback control thus requires active coordination among the neighboring feedback control agents. This problem is not critical when the traffic load is very light, since then all specifications are very easy to meet. It is also not relevant under very heavy load since then the effect of the random perturbations is small compared to the average load, and no green wave solution can exist anyway. However, under intermediate load, when there are a few critical intersections that