On Sample-Efficient Generalized Planning via Learned Transition Models

📰 ArXiv cs.AI

Sample-efficient generalized planning via learned transition models improves solution strategies across planning problems

advanced Published 23 Mar 2026

Action Steps

Learn a transition model from data to improve generalization across planning problems
Use Transformer-based planners, such as PlanGPT and Plansformer, to leverage learned transition models
Apply symbolic abstractions and explicit reasoning over the transition function to further improve planning efficiency
Evaluate the sample efficiency of the learned transition model and refine it as needed

Who Needs to Know This

AI researchers and engineers on a team benefit from this approach as it enables more efficient planning across multiple problems, and software engineers can apply these findings to develop more effective planning systems

Key Insight

💡 Learned transition models can improve the sample efficiency of generalized planning, enabling more effective solution strategies across multiple planning problems

Key Takeaways

Sample-efficient generalized planning via learned transition models improves solution strategies across planning problems

Full Article

Title: On Sample-Efficient Generalized Planning via Learned Transition Models

Abstract:
arXiv:2602.23148v3 Announce Type: replace Abstract: Generalized planning studies the construction of solution strategies that generalize across families of planning problems sharing a common domain model, formally defined by a transition function $\gamma : S \times A \rightarrow S$. Classical approaches achieve such generalization through symbolic abstractions and explicit reasoning over $\gamma$. In contrast, recent Transformer-based planners, such as PlanGPT and Plansformer, largely cast gener

Read full paper → ← Back to Reads