Parallel Strategies for Best-First Generalized Planning

TL;DR

The paper addresses the performance gap in generalized planning by applying parallelization to Best-First Generalized Planning (BFGP). It leverages a frontier-search, Greedy Best-First Search framework with a domain-independent search space of planning programs and uses the heuristic $f(n) = h(n)$ to guide search, enabling non-optimal yet rapid solution discovery. Two shared-memory parallel strategies are proposed and evaluated across nine classical planning domains, achieving substantial speedups (up to around 98x) and revealing tradeoffs between approaches. The results indicate that BFGP is amenable to parallelization and can handle more complex IPC planning problems with future work focusing on node prioritization and asynchronous communication to further enhance performance.

Abstract

In recent years, there has been renewed interest in closing the performance gap between state-of-the-art planning solvers and generalized planning (GP), a research area of AI that studies the automated synthesis of algorithmic-like solutions capable of solving multiple classical planning instances. One of the current advancements has been the introduction of Best-First Generalized Planning (BFGP), a GP algorithm based on a novel solution space that can be explored with heuristic search, one of the foundations of modern planners. This paper evaluates the application of parallel search techniques to BFGP, another critical component in closing the performance gap. We first discuss why BFGP is well suited for parallelization and some of its differentiating characteristics from classical planners. Then, we propose two simple shared-memory parallel strategies with good scaling with the number of cores.