Urban Traffic Control Assisted by AI Planning and Relational Learning

Urban Traffic Control is a key problem for most big cities. An inefficient traffic control system can lead to increased traffic congestions that degrade city quality metrics such as average travel time or city pollution. Most common approaches focus on controlling traffic by appropriately setting traffic lights. Current systems in operation range from static control of traffic light phases to adaptive systems based on numeric models. In this paper, we propose an autonomic approach based on declarative automated planning to generate control plans only when the default behavior should be overridden. Planning is complemented with plan execution control and monitoring, replanning, as well as self-adaptive behavior using Relational Learning. Learning is used to anticipate the appearance of congestions and correctly solve them. Our system outperforms static approaches as well as a planning-based system that recently won a competition on autonomic behavior in Urban Traffic Control.