Intelligent Go-Explore (IGE) integrates the intelligence and internalized human notions of interestingness from giant pretrained foundation models into all stages of the Go-Explore algorithm, enabling foundation model agents to robustly explore in complex environments.
Intelligent Go-Explore: Standing on the Shoulders of Giant Foundation Models; Cong Lu, Shengran Hu, Jeff Clune.
To install, clone the repository and run the following:
conda create -n ige python=3.11
conda activate ige
pip install -r requirements.txt
# BabyAI-Text Instructions
git clone https://github.com/flowersteam/Grounding_LLMs_with_online_RL.git
cd Grounding_LLMs_with_online_RL
pip install blosc; cd babyai-text/babyai; pip install -e .; cd ..
cd gym-minigrid; pip install -e.; cd ..
pip install -e .
# TextWorld Instructions (see https://github.com/microsoft/TextWorld for system libraries)
pip install textworld
- Each environment is one clean file, containing the environment, IGE, baselines, and prompts.
- By default, all runs are enabled, comment out baselines that you don't want.
- Ensure that
OPENAI_API_KEY
is set in the environment variables. - To add a new environment or domain: we recommend cloning a
run.py
file, ensuring the random agent baseline works without problem (no FM needed), and then moving onto IGE.
python3 game_of_24/run.py
python3 babyai/run.py
# Run Coin Collector
python3 textworld/run_coin.py
# Run The Cooking Game
python3 textworld/run_cooking.py
# Run Treasure Hunter
python3 textworld/run_treasure.py
N.B. In order to generate the game files we used in TextWorld found in tw_games
, we used
the text_world/misc/make_*.py
scripts.
If you find this project useful, please consider citing:
@article{lu2024IntelligentGoExplore,
title={{Intelligent Go-Explore}: Standing on the Shoulders of Giant Foundation Models},
author={Lu, Cong, and Hu, Shengran and Clune, Jeff},
journal={arXiv preprint arXiv:2405.15143},
year={2024}
}
Please contact Cong Lu for any queries.