Pytorch Implemtation of Actor-Critic (A3C) and curiosity driven exploration for SuperMarioBros. Idea is to train agent with intrinsic curiosity-based motivation (ICM) when external rewards from environment are sparse.
- Python3.5+
- PyTorch 0.4.0
- OpenAI Gym ==0.10.8
pip3 install -r requirements.txt
First, create a folder (name it "save") in your workspace (os.path). In the save folder, create a .csv file for storing all the log (name it mario_curves.csv). Now you are good to go.
Clone the repository
git clone https://github.com/Ameyapores/Mario
cd Mario
For Actor-critic (A3C), use the following command-
python3 /A3C/main.py
For Curiosity (ICM+A3C), use the following command-
python3 /curiosity/main.py
This would create 4 workers running parallely.
- Dense reward setting
- Sparse reward setting
