A little help understanding virtual enviroments



I am trying to learn a bit about deep learning and especially its application for games. OpenAI gym & Universe seem to be made for that purpose, but I have some questions that I was not been able to find answers to. I am a little confused as to how you define virtual environment. I have seen a list of various flash games used to train so-called agents, but all of the games seemed to be pre-determined.

  • Is there a way to use Universe and give it a unique game? If so, what are you required to give to it?

I have seen it reads input and pixels from the screen (or application window), does that mean you run the agent alongside the application and just make the agent listen to it?