Research map / Large language models / Agents, code and math
One Demo Is All It Takes: Planning Domain Derivation with LLMs from A Single Demonstration
Jinbang Huang, Yixin Xiao, Zhanguang Zhang, Mark Coates, Jianye HAO, Yingxue Zhang
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 |
|---|---|---|---|
| add symbolic PDDL solver instead of using the domain as LLM memory (ablation) | PDDLLM · Nine PyBullet TAMP tasks, overall (50 s limit) | planning success rate (%): 69.9 -> 93.3 | Improved |
| add failure feedback loop to LLMTAMP | LLMTAMP · Nine PyBullet TAMP tasks, overall (50 s limit) | planning success rate (%): 35.7 -> 52.5 | Improved |
| add failure reasons to replanning (LLMTAMP-FR) | LLMTAMP · Nine PyBullet TAMP tasks, overall (50 s limit) | planning success rate (%): 35.7 -> 48.6 | Improved |
| replace GPT-4o with o1 reasoning model | LLMTAMP · Rearrangement/Hanoi/Bridge, extended limit | planning success rate (%): 13.9 -> 61.5 | Improved |
| replace GPT-4o with DeepSeek R1 reasoning model | LLMTAMP · Rearrangement/Hanoi/Bridge, extended limit | planning success rate (%): 13.9 -> 35.9 | Improved |
| use derived PDDL domain instead of o1 reasoning planner | PDDLLM vs o1-TAMP · Rearrangement/Hanoi/Bridge, extended limit | token cost (k): 666 -> 415 | Improved |
| add pick demonstration only (ablation) | PDDLLM · Bridge Building, demonstration | planning success rate (%): 0.0 -> 20.0 with pick and stack added | Hurt |
| add align demonstration to pick and stack (ablation) | PDDLLM · Bridge Building, demonstration | planning success rate (%): 20.0 -> 86.7 | Improved |
| add redundant unstack demonstration (ablation) | PDDLLM · Bridge Building, demonstration | planning success rate (%): 86.7 -> 83.3 | No clear effect |
| use 5 instead of 10 parallel prompts (ablation) | PDDLLM · Unstack | planning success rate (%): 97.8% -> 100% | No clear effect |
| use 5 instead of 10 parallel prompts (ablation) | PDDLLM · Stack | planning success rate (%): 97.8% -> 96.1% | No clear effect |
| shuffle order of dimensions in prompt (ablation) | PDDLLM · Stack | planning success rate (%): 97.8% -> 96.1% | No clear effect |
Closest papers by meaning
- UniDomain: Pretraining a Unified PDDL Domain from Real-World Demonstrations for Generalizable Robot Task Planning (NeurIPS 2025 · similarity 0.91)
- Plan-Seq-Learn: Language Model Guided RL for Solving Long Horizon Robotics Tasks (ICLR 2024 · similarity 0.90)
- Hierarchical Planning for Complex Tasks with Knowledge Graph-RAG and Symbolic Verification (ICML 2025 · similarity 0.89)
- Planning in the Dark: LLM-Symbolic Planning Pipeline Without Experts (AAAI 2025 · similarity 0.89)
- Leveraging Environment Interaction for Automated PDDL Translation and Planning with Large Language Models (NeurIPS 2024 · similarity 0.89)
- Large Language Models as Planning Domain Generators (Student Abstract) (AAAI 2024 · similarity 0.89)
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.
