Research map / Multi-agent systems and games / LLM agents
LLM-Planner: Few-Shot Grounded Planning for Embodied Agents with Large Language Models
Chan Hee Song, Jiaman Wu, Clayton Washington, Brian M Sadler, Wei-Lun Chao, Yu Su
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 kNN retriever for in-context examples (ablation) | LLM-Planner · LOOCV on 100 examples | HLP accuracy: 17.48 -> 40.59 | Improved |
| add logit biases favoring visible objects (ablation) | LLM-Planner · LOOCV on 100 examples | HLP accuracy: 38.10 -> 40.59 | Improved |
| add grounded re-planning to static planning (ablation) | LLM-Planner + HLSM · ALFRED Test Unseen (goal only) | SR: 11.58 -> 13.41 | Improved |
| add grounded re-planning to static planning (ablation) | LLM-Planner + HLSM · ALFRED Test Unseen (step-by-step) | SR: 15.83 -> 16.42 | Improved |
| add grounded re-planning to static planning (ablation) | LLM-Planner + HLSM · ALFRED Valid Unseen (step-by-step) | SR: 14.26 -> 15.36 | Improved |
| use LLM-Planner few-shot instead of retraining HLSM high-level planner on 100 examples | HLSM · ALFRED Test Unseen (step-by-step) | SR: 0.61 -> 16.42 | Improved |
| use generative LLM-Planner instead of SayCan ranking | LLM-Planner · ALFRED Valid Unseen | SR: 9.88 -> 15.36 | Improved |
| use generative planning instead of ranking | LLM-Planner vs SayCan · GPT-3 calls per task | calls: 22 -> 7 | Improved |
| increase training examples from 50 to 100 (ablation) | LLM-Planner · ALFRED Valid Unseen | SR: 10.06 -> 15.36 | Improved |
| increase training examples from 100 to 21k (ablation) | LLM-Planner · ALFRED Valid Unseen | SR: 15.36 -> 17.80 | Improved |
| increase training examples from 10k to 21k | HLSM · ALFRED Valid Unseen | SR: 9.51 -> 18.28 | Improved |
| increase in-context examples beyond 9 (ablation) | LLM-Planner · LOOCV HLP accuracy | HLP accuracy | No clear effect |
Later papers that compared with this method
2 later accepted papers report LLM-Planner as a baseline in their results tables; 2 of them report a better number on most of the results they share with it.
- RoboPARA: Dual-Arm Robot Planning with Parallel Allocation and Recomposition Across Tasks (ICLR 2026 · better on 7 of 7 shared results)
- Multi-Modal Grounded Planning and Efficient Replanning for Learning Embodied Agents with a Few Examples (AAAI 2025 · better on 8 of 8 shared results)
Matched by method name in results tables read by AI. Authors' own tables tend to favour the newer method, and a comparison can be missed when a method is renamed.
Closest papers by meaning
- From Multimodal LLMs to Generalist Embodied Agents: Methods and Lessons (CVPR 2025 · similarity 0.91)
- Large Language Models as Generalizable Policies for Embodied Tasks (ICLR 2024 · similarity 0.90)
- Multi-Modal Grounded Planning and Efficient Replanning for Learning Embodied Agents with a Few Examples (AAAI 2025 · similarity 0.90)
- LoTa-Bench: Benchmarking Language-oriented Task Planners for Embodied Agents (ICLR 2024 · similarity 0.90)
- Embodied Agent Interface: Benchmarking LLMs for Embodied Decision Making (NeurIPS 2024 · similarity 0.90)
- R2C: Mapping Room to Chessboard to Unlock LLM As Low-Level Action Planner (CVPR 2025 · 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.
