First create a new conda enviroment and install the requirements. For best visualization you should have wandb account to see the results and visualizations.
conda create -n jci python=3.8
pip install -r requirements.txt
After go into the torch-ac directory and install torch-ac:
cd torch-ac
pip install -e .
To change the vizdoom environment you need to add both config files to the vizdoom directory:
cp my_way_home_sparse.wad /home/{user}/anaconda3/envs/jci/lib/python3.8/site-packages/vizdoom/scenarios
cp my_way_home_verysparse.wad /home/{user}/anaconda3/envs/jci/lib/python3.8/site-packages/vizdoom/scenarios
cp my_way_home.cfg /home/{user}/anaconda3/envs/jci/lib/python3.8/site-packages/vizdoom/scenarios
Also copy our gymnasium wrapper so you can see the topdown map:
cp base_gymnasium_env.py /home/{user}/anaconda3/envs/jci/lib/python3.8/site-packages/vizdoom/gymnasium_wrapper
Train the agent on the MyWayHome environment with PPO algorithm:
python -m scripts.train --algo ppo --model modelName --no-wandb
``
Note that if you want to log with wandb simply remove the --no-wandb argument. You can change the project name in the scripts/train.py file.
Config for Habitat following...