An Empirical Analysis of Some Heuristic Features for Planning through Local Search and Action Graphs

Abstract

Planning through local search and action graphs is a powerful approach to fully-automated planning which is implemented in the well-known LPG planner. The approach is based on a stochastic local search procedure exploring a space of partial plans and several heuristic features with different possible options. In this paper, we experimentally analyze the most important of them, with the goal of understanding and evaluating their impact on the performance of LPG, and of identifying default settings that work well on a large class of problems. In particular, we analyze several heuristic techniques for (a) evaluating the search neighborhood, (b) defining/restricting the search neighborhood, (c) selecting the next plan flaw to handle, (d) setting the “noise” parameter randomizing the search, and (e) computing reachability information that can be exploited by the heuristic functions used to evaluate the neighborhood elements. Some of these techniques were introduced in previous work on LPG, while others are new. Additional experimental results indicate that the current version of LPG using the identified best heuristic techniques as the default settings is competitive with the winner of the last (2008) International Planning Competition.