Gearing up for FIFA 2022 using RLlib-powered traffic control

As Qatar is gearing up to host the FIFA World Cup 2022, the mobility of the many thousands of people who will attend is a major hurdle for transport authorities. This is an unprecedented challenge considering the fact that all eight stadiums hosting the event are within a 20-mile radius of the city center. We at Qatar Computing Research Institute (QCRI) are working with the local transport authorities to model the traffic demands and plan a traffic control policy targeted at the event. In particular, we have developed: (a) a parallel, congestion-optimized, traffic microsimulator to analyze various what-if scenarios such as closure of a road segment, (b) a prediction model based on Graph Convolutional Networks for traffic in-flow estimation at a traffic light junction, and (c) a multi-agent reinforcement learning framework based on RLlib to design a coordinated traffic light control policy. We will present a brief overview of our work and walk through a demonstration of how we incorporated RLlib to speed up multi-agent learning for coordinated traffic light control.