Research On Multi-Agent Belief Coordination And Action Plan

Multi-Agent System(MAS)is a very active research direction in the field of artificial intelligence.Its goal is to enable software to simulate human cognition and behavior,and it has a strong problem-solving ability.BDI model has flexibility and provides solid foundation for solving the problems in dynamic and complex fields.It expresses the information,motivation and the plans to achieve the desired goals by endowing agent with three mental attributes: beliefs,desires and intentions.In the BDI model,the problem of multi-agent belief coordination and action plan has gradually become a research hotspot.In this paper,the following work is carried out in view of the research status of these two aspects:(1)In multi-agent systems,different agents have different beliefs,and the conflict of beliefs will inevitably lead to action conflict.The rigorous coordination method proposed by Sakama C et al.is only applicable to situations where there is a common belief among agents,but it can't deal with the situation that common beliefs between agents are empty.To solve this problem,we propose a belief coordination method based on Possibilistic Answer Set Programming(PASP).Firstly,according to the different belief sets of agents,the satisfaction degree of the answer set of PASP relative to the belief of agents is calculated based on the weighted quantitative method,so as to weaken some beliefs and obtain a relatively satisfactory consistent solution.Then,a consistent coordination program is established according to the consistent solution,which serves as the background knowledge base for the mutual recognition of agents.Finally,a front-end program for rule preprocessing is added to the DLV solver so that it can deal directly with the rule form of the possibilistic answer set program and implement the algorithm for coordinating the disjoint beliefs of multi-agent.We take the tourist attraction recommendation scenario as an example to illustrate the application of this method in the field of group decision-making.(2)Agents reflect intelligent characteristics through actions.Wang J et al.studied how a agent selects goals and implement actions in an uncertain and fuzzy dynamic environment,but she did not consider the action plan to achieve the goals.Selecting an appropriate action plan to achieve the goal will reduce the possibility of goal failure and improve the performance of the entire agent system.Therefore,we propose a method of action plan under uncertain situation.Firstly,a meta-model containing information such as the current context and agent's preferences is established for various action plans of agents,and the optimal action plan is selected by predicting the implementation effect of various action plans.Secondly,we use the method of practical reasoning to infer actions and revise the goal.Finally,the simulation scenario of military object destruction based on Jadex platform is implemented in this paper to illustrate the practicability of the action planning method.