NES to Neural Network

Hi guys! I would like to ask is it possible to get an input from a NES to a Neural network? I saw a lot of videos in youtube like MarI/O, a Neural network with genetic algorithm that plays mario but the source code is in lua. I would like to build the same but in a different game and in C++. Is it possible to get the Input values there and put it in a C++ neural network?
If you want to do it on a real NES machine it's probably hard to do but not impossible. If you use an Emulator it becomes a lot easier.

Picking an open source emulator is probably the easiest. You probably will have to do some studying to understand how the NES hardware works and also study the game to find out where in memory the game stores the values that you are interested in. I expect it takes a lot of work but could be fun if you are prepared to spend the time.

Mapping the output of the neural network to the NES input controller should be pretty much straight forward to do.
Last edited on
Might be slightly easier if you use an Emulator that has built in scripting support already like MarI/O did with BizHawk.

You wont find something that talks to C++ directly, but you can wrote a small amount of lua then let C++ handle the major stuff by using a library like luabind or something similar. Google can help you with finding NES emulators with lua support.

Of course you would still need to learn all of the memory values etc.
Topic archived. No new replies allowed.