ML-Agent Tanks learn to drive and pick up power ups or die
This is a simple project intended to be used as a fast start for learning and devloping on the Unity MLAgent platform.
The code is built in a way that allows changing many configurtaions with out effecting the basic agent too much.
Setup:
-
Full Project
- Download Git - Make sure you download the Tutorial1 branch
- Open the project
- Open SceneAgents
- Press play
-
As Asset
- Download git - Make sure you download the Tutorial1 branch
- Open new unity package
- install ML agents package from PackageManager
- import assets under tanks into project
- Open SceneAgents
- Press Play
The Project comes with a trained neural network
If you want to train a new network Make sure you have conda installed and that you copy the config directory to your project folder (If you imported the full project it should already be under the root folder) run from conda terminal at project root
mlagents-learn config/YTTanks.yaml --run-id=xxx
you can then see the results in TensorBoard using
tensorboard --logdir=results/
See documents for Tensorboard scalars included or check the driver scalars in the Tensorboard Webpage
This Project uses models from Unity's tanks tutorial The final prefabs have been modefied but the base credit for the models them self belongs to the original creator