This repository contains a simple Deep Q-Network (DQN) agent running in Unity ML Agent that can be used to train and evaluate the result of the training.
I created this for the purpose of learning the basic of Deep Q-Network (DQN).
The DQN is implemented in Python 3 using PyTorch.
Environment
The agent will navigate in a 3D world created in Unity ML where there are yellow bananas and blue bananas.
Goal
The goal is to train an DQN agent to navigate and collect as many yellow bananas as possible in a large, square world.
Rewards
A reward of +1 will be provided for getting a yellow banana
A reward of -1 will be provided for collecting a blue banana
To start training, simply open Navigation.ipynb in Jupyter Notebook and follow the instructions there:
Start Jupyter Notebook
jupyter notebook
Trained model weights is included for quickly running the agent and see the result in Unity ML Agent. Simply skip the training step and run the last step of the Navigation.ipynb