Deep Q-Snake: An Intelligent Agent Mastering the Snake Game with Deep Reinforcement Learning

Debjyoti Ray, Arindam Ghosh, Muneendra Ojha, Krishna Pratap Singh

Abstract: The Deep Q-Network (DQN) has emerged as a robust deep reinforcement learning algorithm capable of learning optimal policies in complex, high-dimensional environments. The Snake game presents unique challenges to an agent's training due to its dynamic and unpredictable nature, making it an intriguing test bed for reinforcement learning techniques. In this paper, we explore the development of “Deep Q-Snake,” an intelligent agent that leverages the power of DQN to master the classic Snake game. By employing DQN, Deep Q-Snake learns to navigate the game grid, collect food, grow in length, and avoid selfcollisions and obstacles. Extensive training and experimentation demonstrate Deep Q-Snake's impressive proficiency in playing the Snake game, achieving high scores, and showcasing strategic gameplay. This work exemplifies DQN's potential to enable intelligent agents to excel in challenging environments and surpass human-level performance when playing the Snake game.