Research On Role-based Collaborative Decision And Planning Under Uncertain Environment For Multi-agent System

Multi-agent system has been widely used in various fields such as defense industry and daily life.The complicated application scenarios put more and more requirements on the capabilities of multi-agent system in coordination tasks.In the uncertain environment,with the incomplete information and the limited communication,multi-agent collaborative decision and planning are facing more and more challenges.Furthermore,it has become a new hotspots in the multi-agent research to introduce the conception of role to design the cooperation framework.It is impressive that with the help of role,the influence of complex conditions on multi-agent collaborative tasks is overcome.Therefore,it is meaningful to study the collaborative decision and planning of role-based multi-agent system.This thesis focues on the collaborative decision problem of multi-agent under the uncertain environment and limited communication.A role-based multi-agent collaborative framework is proposed for multi-agent collaborative decision and planning.The specific contents are listed as following.(1)As the basis of role-based collaborative decision and planning,the multi-agent role assignment problem is discussed.The eigenvector of the role is obtained by analytic hierarchy process to quantify the matching of agent and role.The Hungarian algorithm is used to solve the assignment problem.The simulation results show that the allocation algorithm can obtain the assignment results of roles according to the environmental situation and performance of agents.(2)Aiming at the collaborative decision problem of multi-agent in uncertain and no communication environment,based on partially observable Markov decision process,a rolebased multi-agent implicit collaborative framework is proposed.In order to overcome the lack of teammate status without communication,the framework introduces the state estimation of teammate and updates the estimation based on the observed teammates' action.Based on the framework,a role-based teammate state estimation algorithm is proposed,the validity and accuracy of which are proved theoretically.The simulation results demonstrate the advantages of the framework and the algorithm compared with the traditional decision method.(3)For complex tasks with temporal logic constraints,this paper proposes a role-based linear temporal logic programming based on the finite state transition system and the linear temporal logic programming.The atomic propositions are divided into several subsets according to the tasks of different roles.Each role is planned within its subset,thus the computation load of the problem is reduced.In addition,a role-based knowledge-sharing communication protocol is designed for incomplete information of workspace,which enables the multi-agents to share workspace information based on their roles and realize the ability of agents to adapt flexibly to time-varying tasks.Through the above research,this work provides a feasible solution for collaborative decision and planning of role-based multi-agent system.