摘要： 在智能规划领域中,以往对不确定规划问题的研究主要集中于单个Agent,而对多Agent规划的研究则侧重于确定规划。针对该问题,提出基于多Agent的带权值不确定规划问题,对所求解的强规划解,设计使其所需动作权值总和近似最小的算法。根据基于模型检测的强规划分层方法,对每个Agent进行强规划分层,合并所有Agent的分层信息,并在合并的过程中得到同层状态之间的冲突表。在保证冲突最小的情况下,以最小动作权值优先的贪心方法,求出强规划解。实验结果表明,该算法能较快地求解出使所选择的动作权值总和近似最小的强规划解。

关键词: 多Agent规划, 不确定规划, 强规划解, 模型检测, 动作权值, 智能规划

Abstract: In the field of intelligent planning study,previous studies of nondeterministic planning focus on a single Agent,and research on multi-Agent planning focuses on determining planning.To solve this problem,this paper presents the cost strong planning problem in multi-Agent nondeterministic planning domain,and designs the strong planning algorithm to solve strong planning solution with approximate minimum sum of action cost.Based on model checking of strong hierarchical planning methods to make strong hierarchical planning for each Agent,it merges all Agent’s hierarchical information,and gets the conflicting table of the same level between states in the process of merging at the same time.Under guaranteeing minimum conflicting,this paper uses greedy method for minimum action cost priority to solve a strong planning solution.Experimental results show that the algorithm not only can solve strong planning solution with approximate minimum sum of action cost,but also runs quickly.

Key words: multi-Agent planning, nondeterministic planning, strong planning solution, model checking, action weight, intelligent planning

中图分类号: 

• TP301.6