In transportation networks, two obstacles to real-time control are (1) we don’t know the state, (2) we don’t know the dynamics.

I will illustrate these problems, from experience of working with data from road and transit networks. Ignorance about state sometimes arises simply from noisy observations, but more often it arises from incomplete data collection: missing GPS traces in urban canyons; restricted APIs that only report partial summaries; data trapped in legacy silos. Ignorance about dynamics arises from the complex
interplay between choices made by travellers and the constraints of the transport network: does my control problem involve influencing traveller choices; or coordinating traveller and network actions; or
is it really just about controlling network behaviour?

To address these problems, we have found it useful to reconstruct a "digital replica", using model-based inference to fill in the details of what happens between observations. Visualizations of the digital replica are tremendously useful for gaining insight into what’s going on and how to formulate the right control problem.