Max-pressure intersection control

Max-pressure intersection control models the network as a Markov decision process; queue lengths are the state, traffic signals are the control variables, and stochasticity comes from entering demand and turning ratios. Rather than solve it numerically (which would be computationally intractable due to the curse of dimensionality), max-pressure control proves that an analytical pressure-based adaptive signal control results in a positive recurrent Markov chain if at all possible. Equivalently, network outflow = inflow for max-pressure control if it could be achieved by any signal timing. Furthermore, max-pressure control is decentralized; each intersection optimizes its own timing based on the surrounding vehicle queues.