Novelty Heuristics, Multi-Queue Search, and Portfolios for Numeric Planning

TL;DR

This paper boosts the performance of heuristic search for numeric planning with various powerful techniques orthogonal to improving heuristic informedness: numeric novelty heuristics, the Manhattan distance heuristic, and exploring the use of multi-queue search and portfolios for combining heuristics.

Abstract

Heuristic search is a powerful approach for solving planning problems and numeric planning is no exception. In this paper, we boost the performance of heuristic search for numeric planning with various powerful techniques orthogonal to improving heuristic informedness: numeric novelty heuristics, the Manhattan distance heuristic, and exploring the use of multi-queue search and portfolios for combining heuristics.

Figures (4)

• Figure 1: Legend of domains.
• Figure 2: Left and center: $\text{best }h$, the best SQ GBFS planner for each problem, vs $\text{M}(3h\!\mathop{\mathrm{ {\Vert}}}\limits\!3n)$ on sequential plan length and number of expanded nodes. Right: $\textsc{Patty}$ vs $\text{M}(3h\!\mathop{\mathrm{ {\Vert}}}\limits\!3n)$ on sequential plan length. Points on the top left triangle favour $\text{M}(3h\!\mathop{\mathrm{ {\Vert}}}\limits\!3n)$ and points on the bottom right favour $\text{best }h$ and $\textsc{Patty}$ on the respective plots.
• Figure 3: $h^{\text{md}}$ ($x$-axis) vs. various heuristics ($y$-axis) in terms of plan length and number of nodes expanded during search. Top left points benefit $h^{\text{md}}$ and bottom right points the $y$-axis configuration.
• Figure 4: $h^{\text{add}}_{\left< \text{B},\text{QB} \right>}$ ($x$-axis) vs. $h^{\text{add}}$ and other novelty extensions ($y$-axis) in terms of plan length and number of nodes expanded during search. Top left points benefit $h^{\text{add}}_{\left< \text{B},\text{QB} \right>}$ and bottom right points the $y$-axis configuration.

Theorems & Definitions (2)

• Definition 3.1: Novelty Feature
• Definition 3.2: Novelty Heuristic