The same logic can be used to make an agent for other environments of OpenAI; however, for Atari games such as Breakout or Pac-Man, the observation space is not an array of just four numbers. Instead, it is very large (210 × 160 = 33,600 pixels with 3 RGB values); without some form of quantization, the possible states are infinite and do not yield good results with this simplistic NN. We will resolve this issue using CNN in our deep Q learning recipe.