TRANSITIONING Q LEARNING INTO DEEP LEARNING: A CASE STUDY IN A 2D FIGHTING GAME

Abstract

Reinforcement Learning (RL) is a major advancement in Artificial Intelligence (AI), enabling agents to learn through interaction and feedback instead of direct supervision. In gaming, RL has achieved significant success, especially in dynamic environments like fighting games that demand quick decisions and adaptable strategies. This paper examines the shift from Q-Learning to Deep Q-Learning (DQL) in a 2D fighting game, showing how DQL overcomes Q-Learning’s limits in managing large and complex state spaces. By using neural networks to approximate the Q-function, DQL supports faster learning and more human-like decision-making, proving its strength in developing adaptive and strategic game agents.