The code for the SAC algorithm was built on top of this curl repo This code can be adapated for pixels, but is currently focused on states

Check standard deviation, it had been changed **Uses weight scheme **

• conda environment (curl) taken from the conda_env.yml file
• torch has been downgraded conda install pytorch==1.2.0 torchvision==0.4.0 cudatoolkit=10.0 -c pytorch, and tensorboard has been removed

git clone https://github.com/alec-tschantz/mbmf.git
cd mbmf
conda activate curl
python scripts/sac_script.py

• sac_script.py - train a SAC agent
• mpc_script.py - train an MPC agent
• hybrid_script.py - train a hybrid agent
• test_script.py - test hybrid agent with trained SAC & ensemble model