Improving Hierarchical Planning Performance by the Use of Landmarks

Abstract

In hierarchical planning, landmarks are tasks that occur on any search path leading from the initial plan to a solution. In this work, we present novel domain-independent planning strategies based on such hierarchical landmarks. Our empirical evaluation on four benchmark domains shows that these landmark-aware strategies outperform established search strategies in many cases.

Hybrid Planning Heuristics Based on Task Decomposition Graphs

“…Although the choice of a flaw does not influence correctness or completeness, it heavily influences planning performance. This choice point is one of the major differences to the hierarchical planner used by Elkawkagy et al (2012). That planner orders all modifications and therefore comes without an explicit choice point for flaws.…”

“…Let G = V T , V M , E be a TDG in accordance to Def. 5 given by Elkawkagy et al (2012). Hence, G is a directed AND/OR graph with the following elements: V T is a set of task vertices consisting of ground abstract and primitive tasks that can be obtained by decomposing the initial partial plan.…”

“…SHOP2 (Nau et al 2003) is one of the best-known hierarchical planning systems following that technique. Another approach is to design domain-independent search strategies (Marthi, Russell, and Wolfe 2008;Shivashankar et al 2013;Elkawkagy et al 2012). The hierarchical planning system GoDel (Shivashankar et al 2013), for instance, uses landmarks known from classical planning (Porteous, Sebastia, and Hoffmann 2001) for search guidance.…”

“…The hierarchical planning system GoDel (Shivashankar et al 2013), for instance, uses landmarks known from classical planning (Porteous, Sebastia, and Hoffmann 2001) for search guidance. The hierarchical planning system by Elkawkagy et al (2012) uses hierarchical landmarks to guide its search. These are abstract and primitive tasks, which occur on any refinement process from the initial partial plan to any solution plan.…”

Landmark Generation in HTN Planning

“…However, the gain reported in the empirical evaluation was mainly caused by a presented model reduction, not by the mandatory tasks. In follow-up work Elkawkagy et al (2012) tested how the computed landmark information can be exploited by introducing and evaluating landmark-based search strategies. These search strategies are tailored to the deployed search algorithm, which prioritizes different methods that belong to the same abstract task similar to SHOP (Nau et al 2003;Goldman and Kuter 2019).…”