Learning-based planning systems learn domain-specific knowledge that helps them to solve unseen tasks from the same planning domain. For this purpose they require a diverse set of training instances. A recent proposal for formal specifications of planning domains allows us to exactly characterize which instances are legal for a domain. We automatically generate planning tasks from such formal specifications by means of a translation to answer set programming. We experimentally examine the scalability of the approach and the suitability for learning-based planning, following the setup of the learning track of the International Planning Competition.

Planning domains are an important notion, e.g. when it comes to restricting the input for generalized planning or learning approaches. However, domains as specified in PDDL cannot fully capture the intuitive understanding of a planning domain. We close this semantic gap and propose using PDDL axioms to characterize the (typically infinite) set of legal tasks of a domain. A minor extension makes it possible to express all properties that can be determined in polynomial time. We demonstrate the suitability of the approach on established domains from the International Planning Competition.