Self-Improvement for Fast, High-Quality Plan Generation
Proceedings of the International Conference on Automated Planning and Scheduling · 08.06.2026 · 36 (1) · 601-610
DOI: 10.1609/icaps.v36i1.42878 ↗Kurz gefasst
Fachartikel in der Zeitschrift Proceedings of the International Conference on Automated Planning and Scheduling, online erschienen am 08.06.2026. Verfasst von Robert Gieselmann und 11 weiteren Autor:innen.
Abstract (Originalsprache)
Generative models trained on synthetic plan data are a promising approach to generalized planning. Recent work has focused on finding any valid plan, rather than a high-quality solution. We address the challenge of producing high-quality plans, a computationally hard problem, in sub-exponential time. First, we demonstrate that, given optimal data, a decoder-only transformer can generate high-quality plans for unseen problem instances. Second, we show how to self-improve an initial model trained on sub-optimal data. Each round of self-improvement combines multiple model calls with graph search to generate improved plans, used for model fine-tuning. An experimental study on four domains: Blocksworld, Logistics, Labyrinth, and Sokoban, shows on average a 30% reduction in plan length over the source symbolic planner, with over 80% of plans being optimal, where the optimum is known. Plan quality is further improved by inference-time search. The model's latency scales sub-exponentially in contrast to the satisficing and optimal symbolic planners to which we compare. Together, these results suggest that self-improvement with generative models offers a scalable approach for high-quality plan generation.
Zitieren
APA
Gieselmann, R., Huelsen, H. V., Samson, M., Meyer, M., Piotrowski, D., Radomskyi, O., Okamoto, J., Gojayev, T., Painter, M., Brown, G., Pecora, F., & Wyatt, J. L. (2026). Self-Improvement for Fast, High-Quality Plan Generation. Proceedings of the International Conference on Automated Planning and Scheduling, 36(1), 601-610. https://doi.org/10.1609/icaps.v36i1.42878
Vancouver
Gieselmann R, Huelsen HV, Samson M, Meyer M, Piotrowski D, Radomskyi O, et al. Self-Improvement for Fast, High-Quality Plan Generation. Proceedings of the International Conference on Automated Planning and Scheduling. 2026;36(1):601-610. doi:10.1609/icaps.v36i1.42878
BibTeX
@article{Gieselmann2026,
author = {Gieselmann, Robert and Huelsen, Henrike von and Samson, Mihai and Meyer, Marie-Christine and Piotrowski, Dariusz and Radomskyi, Oleksandr and Okamoto, Justin and Gojayev, Turan and Painter, Michael and Brown, Gavin and Pecora, Federico and Wyatt, Jeremy L.},
title = {Self-Improvement for Fast, High-Quality Plan Generation},
journal = {Proceedings of the International Conference on Automated Planning and Scheduling},
year = {2026},
volume = {36},
number = {1},
pages = {601--610},
doi = {10.1609/icaps.v36i1.42878}
}
EUVIMED