A Local Collaborative Distributed Reinforcement Learning Approach for Resource Allocation in V2X Networks

In this paper, we propose a resource allocation approach for V2X networks based on distributed reinforcement learning with local collaboration. We construct a local collaborative mechanism for sharing information among vehicles. In this model each vehicle is able to obtain the instantaneously information of the environment shared by neighboring vehicles. By adopting proximal policy optimization algorithm, we addressed the issues of the joint allocation of spectrum and transmit power in a distributed manner. The simulation results show that agents can effectively learn an optimized strategy by cooperating with other vehicles in the adjacent area, so as to maximize the ergodic V2I capacity of the whole system while meeting the strict delay limits of the V2V link.