aiqbench analysis
This research addresses a critical reliability gap in LLM-based planning by introducing Localized In-Context Learning (L-ICL), which targets specific constraint violations rather than relying on generic instruction tuning or full trajectory demonstrations. The method significantly improves plan validity across diverse symbolic domains like gridworlds and Sokoban, demonstrating that minimal, step-specific corrections can substantially enhance an LLM's adherence to domain rules.
Localizing and Correcting Errors for LLM-based Planners
arXiv:2602.00276v3 Announce Type: replace Abstract: Large language models (LLMs) have demonstrated strong reasoning capabilities on math and coding, but frequently fail on symbolic classical planning tasks. Our studies, as well as prior work, show that LLM-generated plans routinely violate domain constraints given in their instructions (e.g., walking through walls). To address this failure, we propose iteratively augmenting instructions with Localized In-Context Learning (L-ICL) demonstrations: targeted corrections for specific failing steps. Specifically, L-ICL identifies the first constraint violation in a trace and injects a minimal input-output example giving the correct behavior for the failing step. Our proposed technique of L-ICL is much effective than explicit instructions or traditional ICL, which adds complete problem-solving trajectories, and many other baselines. For example, on an 8x8 gridworld, L-ICL produces valid plans 89% of the time with only 60 training examples, compared to 59% for the best baseline, an increase of 30%. L-ICL also shows dramatic improvements in other domains (gridworld navigation, mazes, Sokoban, and BlocksWorld), and on several LLM architectures.
Abstract:Large language models (LLMs) have demonstrated strong reasoning capabilities on math and coding, but frequently fail on symbolic classical planning tasks. Our studies, as well as prior work, show that LLM-generated plans routinely violate domain constraints given in their instructions (e.g., walking through walls). To address this failure, we propose iteratively augmenting instructions with Localized In-Context Learning (L-ICL) demonstrations: targeted corrections for specific failing steps. Specifically, L-ICL identifies the first constraint violation in a trace and injects a minimal input-output example giving the correct behavior for the failing step. Our proposed technique of L-ICL is much effective than explicit instructions or traditional ICL, which adds complete problem-solving trajectories, and many other baselines. For example, on an 8x8 gridworld, L-ICL produces valid plans 89% of the time with only 60 training examples, compared to 59% for the best baseline, an increase of 30%. L-ICL also shows dramatic improvements in other domains (gridworld navigation, mazes, Sokoban, and BlocksWorld), and on several LLM architectures.
Artificial Intelligence (cs.AI)
arXiv:2602.00276 [cs.AI]
(or arXiv:2602.00276v3 [cs.AI] for this version)
https://doi.org/10.48550/arXiv.2602.00276 Focus to learn more arXiv-issued DOI via DataCite
arXivLabs is a framework that allows collaborators to develop and share new arXiv features directly on our website.
Both individuals and organizations that work with arXivLabs have embraced and accepted our values of openness, community, excellence, and user data privacy. arXiv is committed to these values and only works with partners that adhere to them.
Have an idea for a project that will add value for arXiv's community? Learn more about arXivLabs.
Full text under the publisher's open licence · 253 words
Related coverage
SmellDiffusion: Diffusion-Based Quadruped Navigation with Olfactory Scene Graphs
arXiv · Sep 18, 2026
[AINews] OpenAI reports Navier-Stokes singularity find in 88 hours using Astra-next, roughly 10,000 agents and 130B tokens (>$40M), a contender for second ever Millennium Prize awarded
Latent Space · Sep 9, 2026
CARE: A Lightweight Plug-in Gated Correction and Uncertainty-aware Module for Long-term Time Series Forecasting
arXiv · Oct 9, 2026
ExecuGraph: A Multi-Agent, Execution-Grounded Framework for Reliable Backend Code Synthesis with Large Language Models
arXiv · Jul 24, 2026