A comparison of Different Machine Learning Techniques to Develop the AI of a Virtual Racing Game

Nowadays machine learning (ML) is a field of study that is applied in numerous fields of application: Image classification, Identity fraud detection, Market forecasting, Customer segmentation and others. Another interesting field of study for ML is Video Game. In a virtual environment we can train an Artificial Intelligence (AI) and not script it with thousands of lines of code. A ML-AI could be a Non Player Character that interact with a human player in a friendly or hostile way. It could be an agent that learn how to perform a task in a single player game. The idea of this dissertation is to create a virtual environment with three different AIs trained with a different approach: Reinforcement Learning, Imitation Learning and Curriculum Learning. These agents are trained to compete against a human player in a racing game.