The agent navigates through the square and collects bananas. The goal is to learn how to collect yellow bananas and avoid the purple ones.
- States: 37
- Actions: 4 (forward, back, left, right)
- Rewards: +1 for picking a yellow banana, -1 for choosing a purple banana, 0 otherwise.
Open Navigation.ipynb Jupyter notebook, it also contains a section on how to install the necessary dependencies.
Download the Banana environment (binary file) for your operating system and configure file_name variable (in Navigation.ipynb Jupyter notebook) to point to the location of the binary file.
- Linux: click here
- Mac OSX: click here
- Windows (32-bit): click here
- Windows (64-bit): click here
Install these python packages (could be done via Navigation.ipynb Jupyter notebook itself)
pip --quiet install unityagents
pip --quiet install torch
pip --quiet install numpy
pip --quiet install matplotlib
Open Navigation.ipynb Jupyter notebook and execute each code block to train the agent using DQN.