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. ICLR, 2025.
To install and set up the repository, follow these steps:
-
Clone the repository and create a Python environment:
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 ../gym-minigrid pip install -e . cd .. pip install -e .
-
TextWorld Instructions:
Make sure you have the necessary system libraries (see TextWorld GitHub) and then install:
pip install textworld
- Each environment is encapsulated in its own file, containing the environment definition, IGE implementation, baselines, and prompts.
- By default, all runs are enabled. Comment out any baselines you do not wish to execute.
- Ensure that the
OPENAI_API_KEY
environment variable is set. - Adding a new environment/domain: We recommend cloning an existing
run.py
file, ensuring the random agent baseline works (which does not require foundation models), and then integrating IGE.
Run the Game of 24 environment:
python3 game_of_24/run.py
Run the BabyAI-Text environment:
python3 babyai/run.py
For TextWorld, several game variants are available:
-
Coin Collector:
python3 textworld/run_coin.py
-
The Cooking Game:
python3 textworld/run_cooking.py
-
Treasure Hunter:
python3 textworld/run_treasure.py
Note: To generate the game files used in TextWorld (located in the
tw_games
directory), thetext_world/misc/make_*.py
scripts were utilized.
If you find this project useful, please consider citing our work:
@inproceedings{lu2025intelligent,
title={Intelligent Go-Explore: Standing on the Shoulders of Giant Foundation Models},
author={Cong Lu and Shengran Hu and Jeff Clune},
booktitle={The Thirteenth International Conference on Learning Representations},
year={2025},
url={https://openreview.net/forum?id=apErWGzCAA}
}
For any queries, please contact Cong Lu.