An Empirical Analysis of Some Heuristic Features for Local Search in LPG

LPG is a planner that performed very well in the last International planning competition (2002). The system is based on a stochastic local search procedure, and it incorporates several heuristic features. In this paper we experimentally analyze the most important of them with the goal of understanding and evaluating their impact on the performance of the planner. In particular, we examine three heuristic functions for evaluating the search neighborhood and some settings of the “noise” parameter, that randomizes the next search step for escaping from local minima. Moreover, we present and analyze additional heuristic techniques for restricting the search neighborhood and for selecting the next inconsistency to handle. The experimental results show that the use of such techniques significantly improves the performance of the planner. Introduction The results of the 3rd planning competition (Long & Fox 2003) showed that LPG is an efficient planner for PDDL2.1 domains (Gerevini & Serina 2002; Gerevini, Saetti, & Serina 2003). The system is based on a stochastic local search procedure, called Walkplan, that is similar to the wellknown Walksat procedure for solving SAT problems (Selman, Kautz, & Cohen 1994). As in any local search scheme, the definition of the search neighborhood (the set of possible successor states) and the heuristic function for evaluating its elements are crucial features for the effectiveness of Walkplan. When the number of the elements in the neighborhood is high, its evaluation can be computationally expensive, and a technique for pruning some elements can be very effective. Moreover, in an iterative-repair approach, the strategy to select the next flaw to handle (inconsistency in LPG, unsatisfied clause in Walksat) may also affect the performance of the search. In order to escape from local minima, in Walkplan as in Walksat, if every element in the neighborhood is worse than the current state (according to an heuristic function), then with some probability (called “noise”) an element of the neighborhood is randomly chosen, instead of selecting the best one. In general, the value of the noise can significantly affect the performance of the search. In LPG the noise value can be either statically set by the user, or automatically set Copyright c © 2004, American Association for Artificial Intelligence (www.aaai.org). All rights reserved. to an initial default value that is dynamically changed during search. This paper has two main contributions: • we propose some techniques for effectively restricting the search neighborhood of Walkplan, and for selecting the next inconsistency to handle; • we experimentally analyze the main heuristic features for local search in LPG with the goal of understanding and evaluating their impact on the performance of the planner. In addition to the techniques for neighborhood restriction and inconsistency selection, we analyze three heuristic functions for evaluating the neighborhood elements, that we introduced in previous work (Gerevini & Serina 1999; 2002; Gerevini, Saetti, & Serina 2003), and the noise setting. We focus our analysis on simple STRIPS domains. The second section gives the necessary background on LPG and Walkplan; the third section presents the techniques for the neighborhood restriction and the inconsistency selection; the fourth section presents and discusses the results of our experimental analysis; finally the last section gives the conclusions.