多agent规划领域中的观察信息约简

Observation Reduction in Multi-agent Domain

观察信息约减是近年来不确定规划中的研究热点,但研究集中于单个agent的环境,在多agent规划环境下的研究不足。面对多agent环境下的规划问题,设计了一种用于不确定规划领域中多agent求解协同规划解的ORMAP算法。该算法首先根据基于模型检测的不定规划中的状态分层思想,将问题领域的所有状态进行分层,以此来减少不同的agent的冲突,再利用以最小代价优先的回溯法搜索协同规划解,同时在解的搜索过程中选择最小的观察信息集,使求出的协同规划解在众多符合条件的协同规划解中所需要的观察信息最少或接近最少,这样就达到了信息约简的目的。最后通过实验证明,在考虑了观察信息约简的限制条件后,这种算法的效率较高。

Observation information reduction is a hot area of the uncertainty planning research in recent years,but these researches concentrate on the single agent＇s environment,and the planning problem related to observation information reduction in the multi-agent domain is lack of researching.Confronted with the planning problem in the multi-agent domain,this paper designed an ORMAP algorithm which can find a collaborative planning in the nondeterministic multi-agent domain.At first,the ORMAP algorithm layers all states in the problem domain according to the model-based hierarchical states thought in order to avoid the conflicts between different agents.Then,it searches the collaborative planning solution with the method of the backtracking prior to minimum cost,meanwhile reduces the observation information.At last,a cooperative planning solution can be obtained and it is the one which needs least amount of observation information in all cooperative planning solution to the problem domain,so that it reaches the point.Finally,the experiment shows the efficiency of this algorithm is higher after considering the constraints of the observation information reduction.