智能规划的逻辑编码方式研究

Logical Encoding Methods in Intelligent Planning

逻辑编码方式的设计和实现是基于转换的规划方法有效处理的关键.对几种智能规划方法中的逻辑编码方式予以分析,分别介绍线性编码、基于Graphplan的编码、基于状态的编码、基于动作的编码、基于命题的编码、基于转移的编码、提升的因果编码、基于多值变元的编码、基于有向二元决策图的编码以及基于约束可满足的编码等,并结合国际规划竞赛和相关论文等的实验结论,说明上述编码方式的有效性和可行性,分析该类编码方式在其他领域的应用前景.最后,提出目前智能规划方法中逻辑编码方式研究所面临的挑战、可能的处理方法,以及与之相关的研究热点与趋势.

The design and implementation of logical encoding methods are the key issues of translation based planning methods,which need to translate a given planning problem to a series of other classical solvable problems during planning procedures.All the logical encoding methods need to consider the logical representations and reasonings based on the crossponding propositional logic,first-order logic,multi-value logic,probabilistic logic,modal logic,epistemic logic,or the other adopted non-classical logics.This paper introduces the concrete details of the state-of-the-art logical encoding methods in intelligent planning,which include linear encoding,Graphplan based encoding,state based encoding,action based encoding,proposition based encoding,transition based encoding,lifted casual encoding,multi-value variable based encoding,ordered binary decision diagram based encoding,constraint satisfiabilitiy based encoding and so on.It also introduces the possible needed encoding methods of probabilistics,epistemic properties,modal assumptions,and flexible constraints for planning operations or states of some proposed Abstract planning domain problems,whose formal characteristic expressions are still disputed.After considering experimental results of International Planning Competition and relevant papers,we conclude their corresponding soundness and possibility,and also application prospects in other relevant areas.Finally,we propose the challenges and possible responding methods,and also possible hotspots of them.