Using planning graphs for solving HTN problems Export

@inproceedings{citeulike:875770, author = {Lotem, A. and Nau, D. S. and Hendler, J.}, booktitle = {AAAI/IAAI Proceedings}, citeulike-article-id = {875770}, comment = {This paper describes a way in which the ideas of GraphPlan may be used to increase the efficiency of HTN planning as well as the traditional STRIPS planning. In this algorithm a planning tree is generated stepwise with the planning graph. The planning tree is based on decomposition from the initial task network to possible sequences of primitive tasks that could solve the problem. This is used to restrict the growth of the planning graph from all available operators to only those that may be part of some decomposition for the tasks at hand. Primitive tasks in the planning tree are marked as satisfiable when they appear in the planning graph. Methods in the planning tree are marked as satisfiable when all of their subtasks are satisfiable. Compound tasks in the planning tree are maked as satisfied when at least one of the methods to decompose them is satisfiable. The stepwise generation may stop when the top-level tasks are all satisfiable. This also uses some HTN constraints that are unfamiliar to me, such as guaranteeing that a predicate holds during all time between the execution of one subtask and another. In tests, the algorithm performed significantly better than a baseline HTN implementation, UMCP, and an improved GraphPlan-like implementation, IPP.}, keywords = {htns, planning}, pages = {534--540}, posted-at = {2006-09-27 19:12:30}, priority = {0}, title = {Using planning graphs for solving HTN problems}, year = {1999} }

TY - CONF ID - citeulike:875770 L3 - citeulike-article-id:875770 EP - 540 TI - Using planning graphs for solving HTN problems T2 - AAAI/IAAI Proceedings SP - 534 KW - htns KW - planning N1 - This paper describes a way in which the ideas of GraphPlan may be used to increase the efficiency of HTN planning as well as the traditional STRIPS planning. In this algorithm a planning tree is generated stepwise with the planning graph. The planning tree is based on decomposition from the initial task network to possible sequences of primitive tasks that could solve the problem. This is used to restrict the growth of the planning graph from all available operators to only those that may be part of some decomposition for the tasks at hand. Primitive tasks in the planning tree are marked as satisfiable when they appear in the planning graph. Methods in the planning tree are marked as satisfiable when all of their subtasks are satisfiable. Compound tasks in the planning tree are maked as satisfied when at least one of the methods to decompose them is satisfiable. The stepwise generation may stop when the top-level tasks are all satisfiable. This also uses some HTN constraints that are unfamiliar to me, such as guaranteeing that a predicate holds during all time between the execution of one subtask and another. In tests, the algorithm performed significantly better than a baseline HTN implementation, UMCP, and an improved GraphPlan-like implementation, IPP. AU - Lotem, A AU - Nau, DS AU - Hendler, J PY - 1999/// ER -