In this project, a Reinforcement Learning Deep Q-Network (RL-DQN) was developed and trained to play the classic arcade game Pac-Man. The agent was trained to make decisions based on the visual information from the game screen, which was processed by a Convolutional Neural Network (CNN) model. The CNN was trained to extract useful features from the screen image and pass them as input to the RL-DQN algorithm, which learned through trial-and-error to make optimal decisions to maximize the game score. The project demonstrated strong skills in machine learning, computer vision, and reinforcement learning techniques, and showed the potential of using these methods for training agents to perform complex tasks based on visual input.
Python Ver. - 3.7.15