Skip to content

Read-only version: For security reasons, the Research Information System (FIS) has been temporarily replaced by this read-only version. All public content and full texts remain available; login, editing, submission and statistics are not available. Questions: fis@uni-bamberg.de

Home › Publications › On domain generators for the evaluation of action reversibility in STRIPS

Publication

On domain generators for the evaluation of action reversibility in STRIPS

Schwartz, Tobias; Boockmann, Jan H.; Martin, Leon (2025): On domain generators for the evaluation of action reversibility in STRIPS, in: Bamberg: Otto-Friedrich-Universität, S. 699–726.

Details

Faculty/Chair
Faculty/Professorship
Author
Schwartz, Tobias
Boockmann, Jan H.
Martin, Leon
Publisher Information
Otto-Friedrich-Universität : Bamberg
Year of publication
2025
Source/Other editions
Annals of Mathematics and Artificial Intelligence, Dordrecht [u.a.]: Springer Science + Business Media B.V, 2025, Jg. 93, Nr. 5, S. 699–726, ISSN: 1573-7470, 1012-2443
Year of first publication
2025
Language
English
Licence
URN
Abstract

Robustness is a crucial requirement for the deployment of AI systems in real-world scenarios. In the context of AI planning, the concept of action reversibility, i.e., the ability to undo the effects of an action using a reverse plan, is a promising direction for achieving robust plans. Plans composed exclusively of reversible actions exhibit resilience against goal changes during the execution of the plan. However, the evaluation of action reversibility systems in STRIPS planning presents a challenge, given that standard planning benchmarks are often not suitable. Early experiments using a naive implementation of an action reversibility algorithm show that the available domain generation approach is susceptible to bias. Building on this existing domain generator, we introduce two slight variations that exhibit entirely different search space characteristics. We assess these domain generators using the naive action reversibility implementation and existing ASP implementations, and demonstrate that different generators indeed favor different implementations. As a follow-up to this line of research, we present a generalized domain generator facilitating the creation of domains with diverse search space characteristics. To finally reduce the utilization of contrived generation patterns, we propose another domain generator based on the Barabási-Albert model yielding less rigid domains. Our experiments demonstrate that these new domain generators can produce a variety of domains with diverse search space characteristics, enabling a less biased evaluation of action reversibility systems.

Keywords
STRIPS planning
Action reversibility
Benchmark
PDDL domain generation
Type
Article
URI
Activation date
2025-11-18
Project(s)

Permalink: https://fis.uni-bamberg.de/handle/uniba/111383 · Data as JSON · Cite: BibTeX · RIS

Data as of: 2026-11-15

Restricted read-only access: login, editing and submission are temporarily disabled.