I am currently getting into Deep Learning and would like to set up an environment for training an Artificial Neural Network or NEAT to play simple video games on NES (Mario etc.) and SNES ( Donkey Kong Country etc.), using TensorFlow/TFLearn in Python.

I started off with OpenAI gym environment and there is actually a super-mario environment for gym on github, which I fail to install as Gym-pull is not available anymore and latest gym package doesn't even have scoreboard folder (I am on windows 10, conda environment).

Now, what would be the best way to set up a solid training environment that will be similar to OpenAI gym in terms of simplicity?

Unfortunately OpenAI universe isn't compatible with Windows 10 atm, and I really don't want to get a different setting like ubuntu environment to make it work. I would like to stay in win10.

If someone could guide me for suggested setup or refer me to articles/ documentation where similar things have been done in win10 python env. for NES/SNES, I would be extremely grateful! I assume an emulator with python API (perhaps Nintaco?) is a way to go? How would I then get the 'observation output', I would need to scan the live pixel output of the game, which I am not sure how to do.



  • 1
    $\begingroup$ Could this help? github.com/SerpentAI/SerpentAI $\endgroup$ – Eskapp Apr 30 '18 at 18:52
  • $\begingroup$ Programming issues are off-topic here. Please, read ai.stackexchange.com/help/on-topic for more details and to understand which questions are on-topic here. If you have a programming issue, the best place to ask your question is Stack Overflow. $\endgroup$ – nbro Jun 2 '20 at 23:32
  • $\begingroup$ You may be interested in the following: thekerneltrip.com/reinforcement-learning/nintendo/… which is a work in project, but relies on low level features of the OS (I could not manage to use SerpentAI so I wrote this small article) $\endgroup$ – RUser4512 Jun 4 at 21:22