This project provides a method for the automatic generation of PDDL planning problems, using Graph Neural Networks and Reinforcement Learning.
This work has been partially funded by the Andalusian Collaborative project PYC20 RE 049 UGR and B-TIC-668- UGR20 with FEDER funds.
We want to express our deep gratitude to Masataro Asai, for his suggestion to use NLMs in our work; Simon Stahlberg, for providing the implementation of ACR-GNNs used in a previous version of this work; Mauro Vallati and the rest of authors of [Fawcett et al., 2014], in addition to Sergio Jimenez Celorrio, for their advice on how to measure problem difficulty; Michael Katz, for his advice on how to measure problem diversity; Jiayuan Mao and the rest of authors of [Dong et al., 2019], for their helpful advice on NLMs; and, finally, Christian Muise and the FD community, for their invaluable help on the use of the FD planning system.