I'd like to create an AI for a 2D game involving two players fighting against each other. The map look something like this (The map is a NxN array somehow randomly generated):


Basically the players can look for objects such as weapons located on platforms, shoot at each other to cause damages etc. The output actions are therefore limited to a few such as going up, left, right, down, shooting angle, shooting boolean...

I'm wondering if Reinforcement Learning using a neural network is a good approach to the problem. If so, how should I proceed for the learning phase? Should I force the AI to compete with a weaker version of itself at each iteration? Would it be computationally reasonable to train on a 4Gb GPU? Thanks in advance for your advice !

  • $\begingroup$ Why would you want to train it against a weaker version of itself if you can train it against the current version of itself? $\endgroup$ Dec 11, 2019 at 10:56


You must log in to answer this question.