Reinforcement Learning Based Safety Message Broadcasting in Vehicular Networks

The automatic driving has drawn the widespread attention in both academic and industrial field. The vehicular safety message broadcast plays a significant role in the automatic driving. However, the unique features and transmission requirements of vehicular safety messages are often ignored in the design and analysis of the relevant protocols. This paper focuses on the improvement of safety message broadcasting in vehicular networks with reinforcement learning. In the scenarios with or without road side unit (RSU), we propose the distributed and centralized reinforcement learning based exponential backoff (RLEB) algorithm, respectively. Those vehicles with the dedicated short range communication (DSRC) unit can implement the proposed algorithm with little computation and communication resources. Simulation results show that the proposed algorithms decrease the delays of safety message broadcasting in different scenarios. Furthermore, the centralized proposed algorithm can provide the fairness among vehicles in terms of channel access opportunity.