Traffic Responsive Control of Intersections with Predicted Arrival Times: A Markovian Approach

The dynamic adaptive control of traffic lights can be formulated as a Markov decision problem (MDP). This framework is hardly used, as solving an MDP can be very time-consuming and is only possible for simple infrastructures with a small number of traffic flows. Nevertheless, we show that the MDP framework can be used to construct control policies that approximately minimize the long-run average waiting time at intersections. The MDP-based approach is fast and thus allows real-time use of actual information on traffic conditions, like queue lengths and the arrival times of near-future arrivals. Simulation of an isolated intersection as well as a small network shows that the new policies with arrival information improve pretimed as well as exhaustive control. The new control policies and underlying algorithms scale up well to control networks.