Learning Transition Classifiers for Classical Planning
Dnr:
NAISS 2026/4-1512
Type:
NAISS Small
Principal Investigator:
Farid Musayev
Affiliation:
Linköpings universitet
Start Date:
2026-09-02
End Date:
2027-10-01
Primary Classification:
10210: Artificial Intelligence
Webpage:
Allocation
- Arrhenius Disk at NAISS: 250 GiB
- Arrhenius CPU at NAISS: 10 x 1000 core-h/month
Abstract
This project develops data-driven methods for classical planning. We learn policies as transition classifiers using classical machine learning, and combine them with heuristic search methods for symbolic planning. We aim for models that stay simple and interpretable, while generalizing reliably and scaling to tasks of arbitrary size within a domain.