On-line modelling and planning for urban traffic control

Urban Traffic Control is a key problem for most big cities. Current approaches to handle the city traffic rely on controlling traffic lights. The systems in operation range from static control of traffic light phases to adaptive systems based on numeric models and traffic sensors. Recently, some planning-based approaches have also been proposed. These approaches work at a higher level of abstraction, but have been found to work well if complemented by low-level systems. We have identified two main difficulties for the wide use of planning techniques in this domain: generating the control models is a difficult task; and some algorithms scale poorly. In this paper we present Automated Planning for Traffic Control (APTC), a control system based on Automated Planning, that successfully overcomes these two problems. It combines techniques that continuously: learn an accurate planning model; and also divide the city for distributed reasoning in order to scale to large city networks. Experimental results show that APTC outperforms static approaches as well as other planning-based systems. We also show that the combination of both approaches improves compared with using only one of them.

On-line modelling and planning for urban traffic control Articles