The dimensionality of the state space is 37 and the action space has dimensionality 4. The task in the environment is considered solved in our project whenever the score is greater than or equal to 13.
The project depends on the software packages UnityMLAgents and PyTorch.
Instructions for installing UnityMLAgents can be found here. We consider learning how to develop and build new environments in UnityMLAgents to be outside the scope of this project, but instructions on how to do so are located here.
Instructions for installing PyTorch can be found here.
To train an agent simply run python3.X navigation.py where X is the subversion of python3 you have installed on your system. This will load the Unity environment, train your PyTorch agent, persist the learned model to disk at checkpoint.pth, and create a visualization of your learning curve with matplotlib.