Multi-Agent Deep Reinforcement Learning for Channel Resource Allocation in Vehicular Networks

Vehicular networks rely on periodic broadcast of each vehicle's state information to track its surrounding vehicles and therefore, to predict potential collisions. However, in a scenario of high vehicle density, a large number of vehicles compete to access the shared channel and transmit their beacons, inevitably leading to channel congestion. In order to avoid channel congestion and guarantee the vehicle safety requirements for each vehicle, in this paper, we propose a multi-agent coordinated channel resource allocation method based on multi-agent deep reinforcement learning method. In this method, each vehicle acts as an agent and interacts with communication environments to a learn channel congestion control strategy through updating deep Q-network (DQN). Experimental results show that with the designed reward and training procedure, the proposed method can achieve fair and efficient channel resource allocation for each vehicle under varying traffic conditions.