Performance Analysis of DRL Algorithms in V2V Resource Allocation

The majority of the emerging applications for traffic control and road safety rely on frequently exchanged awareness messages among vehicles. Vehicle-to-Vehicle (V2V) communication enables vehicles to broadcast safety messages, including their positions, velocity, and collision warning information, to reduce road accidents. A dynamically varying environment caused by the increasing mobility of vehicles challenges the usage of conventional resource allocation methods for vehicular communication. Vehicles in the near vicinity tend to use the same resource in a congested case. This leads to the loss of safety messages in vehicular networks. Hence, a reliable method that allocates unique resources to vehicles in a congested and high-mobility environment is required. The aim of the project is to apply Multi-Agent Reinforcement learning with a deep neural network approach for resource allocation. In this method, an autonomous “agent” or a vehicle makes decisions by employing a policy to select the best sub-band and power levels from a pool of shared resources without waiting for global information.