# create a new conda environment
conda create -n rlsound python=3.9
conda activate rlsound
# install impact-driven-rl
git clone [email protected]:Mathieu-Seurin/impact-driven-exploration.git
cd impact-driven-exploration
git checkout soundcur
pip install -r requirements.txt
cd ..
# install MiniGrid
git clone [email protected]:Mathieu-Seurin/minisound.git
cd minisound
pip install -e .
# install torch 1.13
pip install torch==1.13.1+cu117 torchvision==0.14.1+cu117 torchaudio==0.13.1 --extra-index-url https://download.pytorch.org/whl/cu117
python main.py --model vanilla --env MiniGrid-MultiRoom-N2-S4-v0 --total_frames 1000000 --num_actors 3 --savedir outputs/
python main.py --model ride --env MiniGrid-MultiRoom-N7-S4-v0 --total_frames 30000000 --intrinsic_reward_coef 0.1 --entropy_cost 0.0005
python visu_agent.py --expe_path $YOUR_EXPE --env MiniGrid-MultiRoom-N2-S4-v0
Our vanilla RL algorithm is based on Torchbeast, which is an open source implementation of IMPALA.
This code is under the CC-BY-NC 4.0 (Attribution-NonCommercial 4.0 International) license.