Evaluating Action Reversibility in STRIPS

Problem Statement

This thesis will explore the concept of action reversibility in STRIPS planning. Action reversibility generally tries to solve the following problem: Given an action $a$ and a current state $s$, does there exist a plan that returns to state $s$ after executing $a$.
Previous work has developed algorithms for this purpose, but their evaluation remains challenging. We already built a general framework for generating domains that can potentially assess the performance of existing systems.
However, building a complete benchmark is still an important, open challenge!

Goals

• Develop planning domains with explicit and diverse characteristics, such as size, connectivity, and goal distribution.
• Create a benchmark that is more effective for evaluating systems for action reversibility compared to existing ones.
• Consider extensions to algorithms for solving action reversibility, potentially using planners to find reversal plans.

Requirements

Students should have prior knowledge or interest in planning and programming in Python.
Knowledge from the course "algorithms and datastructures" is expected.

Literature

Morak, M., Chrpa, L., Faber, W., & Fišer, D. (2020). On the reversibility of actions in planning. In Proceedings of the International Conference on Principles of Knowledge Representation and Reasoning (Vol. 17, No. 1, pp. 652-661).

Schwartz, T., Boockmann, J. H., & Martin, L. (2022). Towards the Evaluation of Action Reversibility in STRIPS Using Domain Generators. In International Symposium on Foundations of Information and Knowledge Systems (pp. 226-236).