| 摘要: |
| 对智能规划中的常用工具——放松式规划图(relaxed planning graph,简称RPG)的图论性质进行了深入研究.将RPG 中的命题层抽取出来,得到一个不包含任何动作的命题关系图(proposition relation graph,简称PRG),发现PRG 仍具有RPG 的主要规划性质.初步研究结果包括以下4 个方面:初始命题集(initial proposition set,简称IPS)的闭出邻集(close out-neighborhoods,简称CON)是放松式规划可达命题集(relaxed reachable proposition set,简称R-RPS);初始状态命题到目标状态命题的最大距离是规划解长度的合理估计;无圈序指出了对应命题被实现的顺序要求;出度或入度为1 的结点收缩对应规划中构造的宏动作.上述结果中,前两者说明PRG 保留RPG 的主要规划性质,后两者可用于建立目标议程或宏动作提取等领域.还提出与上述结论相关的3 种算法:从RPG 中得到PRG 的算法(复杂性为O(mn2),其中,n 为RPG 的命题数,m 为RPG 的动作数);约简无圈序算法(复杂性为O(n+m),其中,n 为PRG 的结点数,m 为PRG 的边数);宏动作建议算法(复杂性为O(n2),n 为PRG 的结点数). |
| 关键词: 智能规划 放松式规划图 命题关系图 目标议程 宏动作 |
| DOI：10.3724/SP.J.1001.2011.03861 |
| 分类号: |
| 基金项目:国家自然科学基金(61003179, 60903178); 广东工业大学博士科研启动基金(093032); 中央高校基本科研业务费专
项资金(21610305) |
|
| Constructing Goal Agenda and Macro Actions Using Proposition Relation Graphs in Planning |
|
JIANG Zhi-Hua1, RAO Dong-Ning2, JIANG Yun-Fei3, ZHU Hui-Quan4
|
|
1.Department of Computer Science, Ji’nan University, Guangzhou 510632, China;2.Faculty of Computer, Guangdong University of Technology, Guangzhou 510006, China;3.Software Research Institute, School of Information Science and Technology, Sun Yat-Sen University, Guangzhou 510275, China;4.School of Computing, National University of Singapore, Singapore 670527, Singapore
|
| Abstract: |
| This paper focuses on graph properties of relaxed planning graph (RPG), a widely-used tool in
automated planning. When proposition levels are extracted from RPG, and thus, used to build a proposition relation graph (PRG), it is found that PRG keeps primary planning properties in RPG. Preliminary research results include the following four aspects: The close pth out-neighborhoods (CON) of initial proposition set (IPS) is the relaxed reachable proposition set (R-RPS) in planning; the maximum distance from any proposition in initial state to any proposition in goal states is a reasonable estimation of the plan length; acyclic order in graph indicates that some orders that held corresponding propositions are necessary; contraction of in/out cut-vertex means construction of macro-action is currently being planned. The first and second results show PRG keeps planning properties in RPG, and the third and fourth results can be used in goal agenda building and macro-action construction. Three related algorithms are proposed: PRG in RPG finding algorithm, an O(mn2/4) (n is the number of propositions in RPG, m is the number of actions in RPG) algorithm; acyclic order reduction algorithm, an O(n+m) (n is the number of nodes in PRG, m is the number of edges in PRG) algorithm; macro action suggestion algorithm, an O(n2) (n is the number of nodes in PRG) algorithm. |
| Key words: AI planning relaxed planning graph proposition relation graph goal agenda macro action |
