Machine learning network learning to play sonic the hedgehog. With Openai gym environment, both single core and multicore learning using a feedforwrad neuralnetwork.
signal core learning: https://www.youtube.com/watch?v=ZV-5VjhPaY8&t=20s
Muilt Core learning: https://www.youtube.com/watch?v=oKpx9jT7GJk
I started this project around 6 months ago and I've been working on it off and on. I started this awesome projects on the basis of a school assignment and I was so fascinated by the idea of having a neural network learning with no prior knowledge of how to play a video game. But as with everything it took time and a lot of problem solving I followed a couple tutorials on YouTube and lots of reading and posting questions on github. At first I tried to install open AI on my Mac I got decently through the installing process after I got too many fatal errors and software hasn't been installed properly. Until I came to the conclusion that I have to restart on a Windows machine. I installed the Linux environment and X window on a Windows 10 machine that what work pretty well but because it was still running on a Windows machine it couldn't open up graphics using x windows. So I finally decided to leave that and to move to a pure Linux machine and that worked out pretty well and it's what I ended up using to complete this project.
My end goal for this project was to demonstrate and educate people about the capability that AI has to complete and solve problems. I thought what's the best way to prove this then having an AI play a video game which is full of problems and riddles that you have to solve, how do I jump over this box, how do I not get attacked by this. If a self taught neat network can complete a video game surely it can start to learn and understand human orientated problems like in a workplace for example.
Quick tip don't use Mac don't use Windows use Linux
I recorded every command and library that I installed that helped to resolve all the issues in a document called "Complete install list for gym retro.txt"