Per-Domain Generalizing Policies: On Learning Efficient and Robust Q-Value Functions

Abstract

Learning per-domain generalizing policies is a key challenge in learning for planning. Standard approaches learn state-value functions represented as graph neural networks using supervised learning on optimal plans generated by a teacher planner. In this work, we advocate for learning Q-value functions instead. Such policies are drastically cheaper to evaluate for a given state, as they need to process only the current state rather than every successor. Surprisingly, vanilla supervised learning of Q-values performs poorly as it does not learn to distinguish between the actions taken and those not taken by the teacher. We address this by using regularization terms that enforce this distinction, resulting in Q-value policies that consistently outperform state-value policies across a range of 10 domains and are competitive with the planner LAMA-first.

@inproceedings{mullerPerDomainGeneralizingPoliciesOn2026,
 author = {Müller, Nicola J. and Oster, Moritz and Valera, Isabel and Hoffmann, Jörg and Gros, Timo P.},
 booktitle = {Proceedings of the Thirty-Sixth International Conference on Automated Planning and Scheduling. International Conference on Automated Planning and Scheduling (ICAPS-2026), June 27 - July 2, Dublin, Ireland},
 doi = {10.1609/icaps.v36i1.42889},
 isbn = {ISSN 2334-0843},
 month = {6},
 number = {1},
 pages = {703-707},
 publisher = {AAAI Press},
 title = {Per-Domain Generalizing Policies: On Learning Efficient and Robust Q-Value Functions},
 url = {https://ojs.aaai.org/index.php/ICAPS/article/view/42889},
 volume = {36},
 year = {2026}
}