Research On Multi-agent Cooperative Pursuit Method Based On Game Theory

The multi-agent cooperative hunting problem studies a hunting team composed of multiple agents to cooperate with each other to hunt down another group of mobile agents.During the hunt-and-run process,the Agent hunting team needs to coordinate and cooperate with each other to complete the hunting task.The multi-agent system is the cutting-edge direction in the field of distributed artificial intelligence.The problem is a typical problem in multi-agent cooperation research.The key technologies involved are also widely used in military and industrial fields,such as battlefield search and rescue,capture operations,cooperative encirclement of agent forces,capture of intruders,and air combat field.In the multi-agent cooperative hunting problem,under the condition that the escaper and the hunter have the learning ability,both the escaper and the hunter will receive the influence of the other party’s choice of their behavior strategy,and the hunting team will also Behavior affects other hunters.Game theory provides a good mathematical tool for this mutual influence decision-making problem.At the same time,the method of reinforcement learning for the participating agents to maintain teamwork and learning in the process of performing tasks the target execution strategy and update your own action strategy provide a learning model.This article uses game theory and reinforcement learning to study the problem of multi-agent cooperative hunting.The specific content is as follows:(1)First,the research status of multi-agent cooperative hunting at home and abroad is summarized.Introduce the current research status of multi-agent cooperative hunting,the purpose and significance of the multi-agent hunting problem,analyze the results obtained by your own research,andpoint out the shortcomings;secondly,introduce the basic theory of multi-agent system and multi-agent cooperation,and this article Study the game theory and the related theorems of reinforcement learning required for multi-agent cooperative pursuit.(2)The problem of single-runner hunting in a perceptible environment with obstacles is studied.Aiming at the escapees with learning ability,a multi-agent cooperative pursuit algorithm based on game theory and Q-learning is proposed,and a method of virtual action is used to solve the problem of multiple balanced choices,which effectively ensures the successful completion of the task.Simulation experiments verify the adaptability and stability of the multi-Agent team in pursuit of a single escaper in the obstacle and perceptible environment of the proposed algorithm.(3)The problem of chasing multiple fugitives under obstacles is studied.Aiming at the problem of multiple fugitives with learning ability,and the fugitives with learning ability,the fugitives can generate arrest behaviors through mutual cooperation.This paper proposes a method of multi-agent cooperative pursuit of multiple targets.This method is based on prediction— The planning-executing multi-agent hunting algorithm idea decomposes the multi-agent hunting problem into dynamic multi-agent motion state prediction and dynamic multi-agent target trajectory learning;team collaboration and virtual agent decision-making between escaping agents;hunting down teams and target escapers The game is solved by studying three sub-problems.The algorithm is verified in a simulation platform.The results show that the algorithm has fast convergence and high feasibility.(4)Based on the Microsoft Visual Studio Community 2013 development platform,this article designs and implements a multi-agent collaborative pursuit simulation platform.A mature game algorithm and reinforcement learning algorithm are built into the platform,and the algorithm proposed in this article is available for users to choose.At the same time,it provides users with a visual operation interface in theplatform,which is convenient for users to choose the visual terrain and the number of Agents.The platform is used to carry out simulation experiments of hunting problems to verify the feasibility of the algorithm proposed in this paper.(5)Finally,summarize the main work of the paper and explain the next work.