To create your virtual environment, you can run the following commands

conda create -n IL python=3.8
conda activate IL
pip3 install torch torchvision torchaudio
pip install stable-baselines3[extra]
pip install gymnasium
pip install tensorboard
pip install rl_zoo3
pip install pybullet
pip install pybullet_envs_gymnasium
# for IQ-learn
pip install wandb
pip install hydra-core --upgrade
pip install tensorboardX
pip install termcolor
<!-- pip install 'shimmy>=0.2.1' -->

To run our method, you should first train an IQ-learn model by

cd baselines/IQ-Learn/iq_learn
python train_iq.py env=cheetah agent=sac expert.demos=1 method.loss=value method.regularize=True agent.actor_lr=3e-05 seed=0

For other environment, IQ-learn training scripts can be found in 'baselines/IQ-Learn/iq_learn/scripts'; However, for mujoco tasks, iq-learn use the original mujoco env but we use pybullet env, so you may need to alter the config file in 'baselines/IQ-Learn/iq_learn/conf/env' and change the env name to pybullet version (I've done so for halfcheetah and hopper).

After training IQ-learn agent, you can run our method (I name it Coherent Q-function Imitation Learning (CQIL) for now) by

python train_cqil.py

Run behavior cloning and CSIL by

python train_bc.py
python train_csil.py

Note: Current code only support continuous env in mujoco.