Routing for crowd management in smart cities: A deep reinforcement learning perspective

The concept of smart city has been flourishing based on the prosperous development of various advanced technologies: Mobile edge computing (MEC), ultra-dense networking, and software defined networking. However, it becomes increasingly complicated to design routing strategies to meet the stringent and ever changing network requirements due to the dynamic distribution of the crowd in different sectors of smart cities. To alleviate the network congestion and balance the network load for supporting smart city services with dramatic disparities, we design a deep-reinforcement-learning-based smart routing algorithm to make the distributed computing and communication infrastructure thoroughly viable while simultaneously satisfying the latency constraints of service requests from the crowd. Besides the proposed algorithm, extensive numerical results are also presented to validate its efficacy.