A Planning-Based Approach for Generating Planning Problems

Most of the research in Automated Planning relies on the evaluation of different techniques over a set of benchmarks. The generation of planning tasks for these benchmarks is done using generators coded ad-hoc. Instead, we propose an approach for generating planning problems automatically given the domain definition and some declarative semantics-related information provided by the user. The approach consists of modelling the task of generating planning problems also as a planning problem. The main contribution of this work is that the generation of planning problems is partially handled in a domain-independent way, which leads to a saving of time and effort for researchers. Additionally, the declarative input to the generator facilitates the modification of its behavior. This is a feature of interest for generating different problem distributions.