CopyConda

    Reinforcement Learning vs Imitation Learning Snake Game

    Machine LearningPythonReinforcement LearningImitation LearningPyTorchDeep Q-Networks

    Motivation

    In the realm of Machine Learning, the Snake Game has been a target for multiple researchers attempting to make a computer beat the game. Often, the paradigm that is used for this is Reinforcement Learning. For this project, Reinforcement Learning was compared with Imitation Learning to compare the results.

    Results

    By the end, a Deep Q-Network (DQN) and an Imitation Learning pipeline were developed and compared against each other. This project strengthened the use of the Reinforcement Learning paradigm over traditional Machine Learning. While the Reinforcement Learning (DQN) model was able to achieve results over 20 points consistently, the Imitation Learning model struggled to achieve even 2 points in a single game.

    Presentation

    Videos

    Report