Research map / Large language models / Agents, code and math
Automated Repair of Totally-Ordered Hierarchical Task Network Domains via Context-Free Grammars with Large Language Model Support
Daniel Lutalo, Pascal Bercher
Read the paperCheck your idea against 137,217 papers
AI-extracted. The measured effects below were read by AI from the paper's full text (no summary yet). Numbers and claims can be wrong or missing context, so check the paper before you cite or build on it.
Measured effects reported in the paper
| What was changed | Where | Measured | Effect |
|---|---|---|---|
| replace symbolic re-encoding with LLM-based CFG repair (o4-mini) | TO-HTN domain repair · TO-HTN repair benchmark (156 instances) | solved instances: 76 -> 145 | Improved |
| use open-weight gpt-oss-120b instead of o4-mini | LLM-based repair · TO-HTN repair benchmark | solved instances: 145 -> 140 | No clear effect |
| replace optimal encoding with suboptimal greedy encoding | symbolic repair encoding · TO-HTN repair benchmark | solved instances: 57 -> 76 | Improved |
| use LLM repair instead of suboptimal encoding on Monroe (FO) | TO-HTN domain repair · Monroe (FO) | solved instances: 0 -> 17 | Improved |
| use LLM repair instead of suboptimal encoding on Childsnack | TO-HTN domain repair · Childsnack | solved instances: 10 -> 9 | Hurt |
| use LLM repair instead of optimal encoding in terms of repair cost | TO-HTN domain repair · 57 instances with known optimal cost | repair cost: optimal in 42 of 57 cases; 6% higher average cost | Mixed |
| use greedy suboptimal encoding instead of optimal encoding for repair cost | symbolic repair encoding · 57 instances with known optimal cost | repair cost: 44% higher average cost | Hurt |
| increase plan length (plans longer than about 40 steps) | LLM-based repair · TO-HTN repair benchmark | failure rate: plan length p = 0.015; token count p = 0.052 | Hurt |
| add structural filtering and CFG pruning before prompting the LLM | LLM-based repair · TO-HTN repair benchmark | prompt size and coverage | Improved |
| mask all action and method names in the prompt | LLM-based repair · TO-HTN repair benchmark | data leakage risk | Improved |
Closest papers by meaning
- RefactorBench: Evaluating Stateful Reasoning in Language Agents Through Code (ICLR 2025 · similarity 0.86)
- ReCAP: Recursive Context-Aware Reasoning and Planning for Large Language Model Agents (NeurIPS 2025 · similarity 0.86)
- AuPair: Golden Example Pairs for Code Repair (ICML 2025 · similarity 0.85)
- Improving Large Language Model Planning with Action Sequence Similarity (ICLR 2025 · similarity 0.85)
- RePO-VLA: Recovery-Driven Policy Optimization for Vision-Language-Action Models (NeurIPS 2026 · similarity 0.85)
- Reinforced Context Order Recovery for Adaptive Reasoning and Planning (NeurIPS 2025 · similarity 0.84)
Writing a paper in this area? scime draft is a free LaTeX editor with AI included, real-time co-authors and a check of your references and venue rules before you submit. It also imports OpenAI Prism projects.
