Working-sets in automated planning: Finding better performing sets of actions in established domains
Translated title
Working-sets i automaseret planlægning
Author
Jakobsgaard, Søren Drejer
Term
4. term
Education
Publication year
2024
Submitted on
2024-06-09
Pages
15
Abstract
Automated planning is a complex issue and multiple avenues are researched to help solve planning problems. Composite actions like metaactions are one such method. These meta-actions functions as shortcuts making planners skip steps they would otherwise be deciding the next action. Adding these extra actions through, increases branching factor so meta-actions can’t be added haphazardly. This paper explores the idea of removing actions that are made redundant from meta-actions, thereby decreasing the overall branching factor and hopefully improve performance of planners. First some preliminaries actions are taken to ensure that removing the actions does not result in sets of actions that can not solve problems. Afterwards, an AI tool called SMAC is used to select sets of actions based on different metrics. These selected sets are then sent to a cluster and tested alongside other sets to compare results. The results show there is broadly a gain in planning speed and coverage when removing redundant actions across domains. But each planning domain is very distinct, so no completely common pattern to what feature makes an action more or less suited to be included/excluded could be determined.
Documents