Research And Design Of Multi-agent Hierachical Communication Coop-Eration Method

In the intelligent transportation system,vehicles can be abstracted as agents.Through V2X communication between vehicles,multi-agent joint perception and collaborative computing are realized.The collaborative computing of multi-agent is very dependent on communication efficiency.At present,vehicle groups do not consider the different coordination requirements under different complexity traffic scenarios in the process of coordination,resulting in the competition or unsaturated use of communication resources.To solve this problem,this paper studies the cooperation quality and communication cost of multi-agent in different complexity scenarios,and proposes a multi-agent hierarchical communication cooperation framework.Firstly,this paper proposes a multi-agent hierarchical communication and cooperation method based on vehicle traffic sub-target.This method adopts a two-level decision network.The upper network communicates with each other and is responsible for setting personalized traffic sub-goals for each vehicle.The lower network is responsible for controlling the vehicle’s actions to achieve the traffic sub-goals set by the upper layer to complete the driving task.Because the collaboration demand of the agent only occurs in the upper layer,and the upper layer makes decisions on the time scale of the sub-target,the communication demand in the process of the agent collaboration is reduced.In this paper,a traffic intersection simulation environment with multi-vehicle self-organization is established and verified by experiments.The experimental results show that the hierarchical communication cooperation method uses less communication to achieve a traffic success rate of more than 10%by setting a reasonable cooperation granularity for the upper layer network for the multi-vehicle swarm intelligence cooperation scenario with the same complexity.In order to further improve the flexibility of multi-agent swarm intelligence collaboration and realize the adaptive adjustment of collaboration granularity in dynamic complexity scenarios,a three-level communication collaboration method based on complexity perception is designed.This method adds the collaborative control network as the third layer above the two-layer communication collaboration architecture,captures the complexity information of the environment,and sets the collaboration granularity under the current scenario for the lower two levels.To solve the problem of insufficient feedback information for end-to-end update,the top-level network introduces additional reward signals and adds regular items as communication resource expenditure,so as to ensure the balance between cooperation efficiency and communication resource consumption.The experimental results show that this method can learn adaptive cooperation granularity in different complexity scenarios,realize the joint optimization of communication resources and cooperation quality,and reduce the communication cost by 12%on average while ensuring the cooperation quality.