Code for LLM-Planner.
Check project website for an overview and a demo.
- Dec 23: LLM-Planner with support for an oracle low-level planner with a new easy-setup framework with ALFOWLRD backbone. We hope this codebase can serve as a foundation for building LLM or LMM based methods with ALFRED.
- High level planner
- KNN dataset
- KNN retriever
- Low level planner
- Oracle low level planner
- HLSM low-evel planner
- Fine-grained control over visualization
- Support for non-OpenAI foundation models
Clone repo:
git clone https://github.com/OSU-NLP-Group/LLM-Planner
cd LLM-Planner
export ALFWORLD_DATA="$(pwd)/alfworld/data"
Install requirements:
# Conda or Python enviornment recommended
# Install requirements for the AI2Thor simulator and ALFRED
cd alfworld
pip install .
# Install requirements for LLM-Planner
cd ../src
pip install -r requirements.txt
Download data:
cd ../alfworld
alfworld-download
Sanity check on AI2Thor simulator
python scripts/check_thor.py
# This should return successful, if not your AI2Thor simulator is not set up correctly.
Start evaluation with GPT-4
export OPENAI_KEY=<Your OpenAI Key>
cd ../src
python run_eval.py --config gpt4_base_config.yaml
Coming soon.
Tested on:
- Mac M1
- Ubuntu 18.04
If you find this code useful, please consider citing our paper:
@InProceedings{song2023llmplanner,
author = {Song, Chan Hee and Wu, Jiaman and Washington, Clayton and Sadler, Brian M. and Chao, Wei-Lun and Su, Yu},
title = {LLM-Planner: Few-Shot Grounded Planning for Embodied Agents with Large Language Models},
booktitle = {Proceedings of the IEEE/CVF International Conference on Computer Vision (ICCV)},
month = {October},
year = {2023},
}
We thank the authors of ALFWORLD for releasing their code.
- LLM-Planner - MIT License
- ALFWorld - MIT License
Questions or issues? File an issue or contact Luke Song