Resources

Information

Using Unity, I created a custom made FNAF 1 environment that I can extract information from.

Then, using [socket], I connect this environment to my RL agent in python.

This connection allows me to send data between the env and the agent, such as env state and playing actions.

The AI model is a D3QN with prioritized replay experience.

Michael's Description

This was the first reinforcement project I did. After learning all the math and stuff, I watched an intro to RL with a snake game. I was like, "damn, I wonder if I can make it play FNAF"

And then I did. It sucked a lot because I had no idea what I was doing (the input layer was wrong for 2 weeks). After actually reading papers and how to implement one, I was able to fix it

and make it a D3QN model instead. It tends to waste power very often, and can only survive a max animatronic level of 7 (all of them at level 7)

Visuals

NOTE: i made my conda terminal green during this; i thought it made me look cool lmao