Post 3: Growing Pains with Unity ML Agents

Unity, as most people in the tech industry know, is a game engine. You can use it to make games of all shapes and sizes, 2D and 3D, with as little or as much complexity as you want. In last week’s post, I mentioned our team got the Machine Learning (ML) Breakout project. Well I’ve spent some time this week learning some basics of Unity, and watching some YouTube videos on ML Agents in Unity.

I’ve got to say I feel overwhelmed. Machine Learning in a general sense is still new to me, as is Unity. Put them together and that is “new x new = new squared” if we were to look at it from a math standpoint. Nevertheless, we move forward.

This week I was able to get ML Agents running a demo example, one that was already totally set up and just trying to run it. I then tried to follow along a video example of another project and apply it to a basic Breakout like game, and failed miserably. Technically ML Agents is running, but it is not handling targets and rewards properly, and the controls of the paddle go outside the bounds of the game. Because of this, the model is not learning anything useful, nor is it appearing to improve.

I think what I am trying to say, is that I need to go back to the drawing board. I need to start with some more simple implementations of ML Agents, and work up to applying it to a Breakout game. Maybe then I will understand what I’m missing on how to set it up. That’s a good spot to leave things for now, next week I hope to have better news.

Happy coding!

Print Friendly, PDF & Email






Leave a Reply

Your email address will not be published. Required fields are marked *