Towards Canonical and Minimal Solutions in a Constraint-Based Plan-Space Planner
Year: 2025
Type: book-chapter
Abstract: Temporal planning is the problem of selecting and scheduling a set of actions in order to achieve a distant objective from an initial state. Constraint Programming (CP) has historically played an important role in many automated temporal planners based on the Plan-Space Planning (PSP) algorithm. In a nutshell, PSP would drive the search for a plan by selecting which actions to add and how to resolve conflicts, with an embedded CP engine in charge of ensuring the coherence of the plan. In this paper, we consider an existing encoding of temporal planning into constraint programming that adheres to a plan-space approach, but encodes the decision of the PSP algorithm as decision variables of the CSP, thus letting the CP engine entirely drive the search for a solution plan. Analyzing the encoding, we show that the induced search space contains many symmetries that would be absent from the original PSP algorithm. Another caveat of the encoding is that it allows for non-minimal solutions, i.e., solutions in which some actions may be removed without invalidating it. We propose to extend the encoding with additional symmetry breaking constraints to enforce canonicity of the considered plans. The encoding is further improved with additional constraints to limit the presence of non-minimal plans. We show the extended encoding to very substantially improve the performance of the original version and to be competitive with state-of-the-art state-space temporal planners on the considered benchmarks.
DOI: 10.3233/faia251373
Authors: Arthur Bit-Monnot, Roland Godet
Institutions: Centre National de la Recherche Scientifique, Laboratoire d'Analyse et d'Architecture des Systèmes, Institut National des Sciences Appliquées de Toulouse, Université de Toulouse, Office National d'Études et de Recherches Aérospatiales, Département Traitement de l’Information et Systèmes
Language: English
Subfield: Artificial Intelligence
Field: Computer Science
Domain: Physical Sciences
Open Access status: hybrid
Funder: Agence Nationale de la Recherche
OpenAlex ID: W4414879616
