Research And Implementation Of Efficient Multi-agent Communication Method In Vehicle-road Collaborative Environment

Intelligent transportation system(ITS)is an important research direction to improve traffic efficiency,and it makes up for the limited perception and computing ability of single intelligent vehicle through edge perception and computing ability.In the intelligent vehicle-road cooperation system,the cooperative performance depends on the cooperative communication quality.In order to improve the quality of cooperative communication,this paper proposes a vehicle-road multiagent cooperation framework,which establishes the information space agent topology based on the positional relationship of collaboration entities in physical space and cooperation requirements,and proposes corresponding efficient communication methods for the problems faced by communication under different cooperation scales.In small-scale cooperation,multiple intelligent agents can share information based on fully connected communication.However,the limitation of communication resources is not considered in the existing related research,which leads to communication carrying capacity is not effectively considered in the process of vehicle-road coordination,and communication resources are wasted with invalid redundant information.Aiming at the fully connected collaborative communication under the condition of limited communication resources,this paper firstly compresses the information vector to solve the bandwidth limitation problem of multi-agent communication;then,it extracts multi-angle information based on the multi-head attention mechanism for the agent according to the current state of the agent to improve the quality of vehicle collaborative decisions;finally,it makes sending decision according to the evaluation of sending information,the occupation of communication resources by invalid information is reduced.Simulation tests in multijunction traffic network show that the proposed method can improve the effectiveness of multi-intelligent vehicle cooperative communication under limited communication resources and reduce the occupation of communication resources.In large-scale collaboration,subject to the constraints of communication distance,multi-vehicle collaboration entities need to build a graph communication topology based on spatial location to meet the needs of large-scale collaboration.However,the topological characteristics of cooperative information and the limitation of communication resources are not considered in the existing related research,which cannot provide a reliable guarantee for the cooperative performance.For collaborative communication in large-scale graph topology scenarios,this paper proposes a multi-agent efficient graph convolution communication method.In this method,firstly,information bottleneck is used to realize the initial compact coding of agent observation and intention information to ensure the initial coding is compact;then,based on the graph information bottleneck and the neighbor sampling mechanism,the graph convolution calculation of the agent’s collaborative content is realized.The agents are abstracted into graph nodes,and the communication content between agents is improved from original data to graph convolution data,thereby reducing data redundancy in the communication process.The simulation test in the multi-intersection traffic network shows that integrating the multi-agent cooperative communication process and the graph convolution calculation process to construct a multi-agent efficient graph convolution communication method can effectively guarantee the large-scale vehicleroad cooperative communication performance and improve the communication effectiveness and robustness.