Research In Multi-agent Cooperation Algorithm Based On Featured Message Communication

Multi-agent systems will be one of the most important agent systems.Multi-agent system applications include self-driving automobiles,robot swarms,training systems of sports,etc.Cooperation is a very important interactive method between agents in a multi-agent system.Among all modern methods,cooperation with communication is the most effective method with the least amount of calculation.However,the existing methods have problems such as excessive communication volume and insufficient stability.In order to reduce the amount of communication between agents,improve the overall stability of the system,and then improve the overall success rate of the multi-agent cooperation system,this paper proposes a set of multi-agent cooperation systems based on characteristic information communication.Agents in this system can communicate through the characterized information,thereby reducing the amount of communication and improving the overall system efficiency.The main research results of the paper include:1)Designed and implemented a multi-agent cooperation system based on characteristic information communication.Through the directional transmission of characteristic information,the communication volume and the number of communication times are reduced,and the overall system cooperation efficiency is improved.In addition,the system improves the computing efficiency of each module by dividing the agent architecture into behavior modules and communication modules;2)Proposed a feature-based information generation algorithm based on a distiller.By extracting features from the agent information,the goal is to reduce the amount of data communication between agents and improve the computing stability of the agents;3)Proposed a multi-agent communication target decision algorithm based on a multi-labeled fully connected neural network.The algorithm automatically selects the appropriate communication target through the probability table calculated by the neural network to achieve the goal of reducing the number of communication times and improving communication efficiency;4)Designed and implemented several experimental environments that can run stably and conducted experiments.Through experimental verification,the agent framework and multi-agent communication module proposed by the paper perform the best in experiments and the most stable compared to other traditional methods.