Hands-On Neuroevolution with Python

Game screen mapping into actions

Deep ANNs can be trained to play Atari games if they can directly map the pixels on the screen to a system to control the game. This means that our algorithm must read the game screen and decide what game action to take to get the highest game score possible.

This task can be divided into two logical subtasks:

• The image analysis task, which encodes the state of the current game situation on the screen, including the game character's position, obstacles, and adversaries
• The RL training task, which is used to train the Q-value approximation ANN to build the correct mapping between the specific game state and actions to be performed

Convolutional Neural Networks (CNNs) are commonly used in tasks related to ...