Multi-Algorithm Approach to Snake Game: A Comprehensive Study of Minimax, Reinforcement Learning, and Heuristic Search Methods

Zhiyuan Peng, Xin Yin, Rui Qian, Manning Qian, You Wu, Chenhao Ying, Yuan Luo

Keywords: Snake game, artificial intelligence, minimax, reinforcement learning, heuristic search, algorithm comparison, evaluation framework, game AI

TL;DR: We compare minimax, reinforcement learning, and heuristic algorithms for Snake Game, showing RL scores highest, minimax is most consistent, and heuristics are fastest in real time.

Abstract: The Snake game serves as an exemplar of artificial intelligence challenges, encompassing path planning, collision avoidance, and strategic decision-making in a dynamic environment. We present a rigorous comparative analysis of diverse algorithmic approaches: minimax with alpha-beta pruning, advanced reinforcement learning methods (DQN, A3C, PPO), and heuristic search algorithms (A*, greedy best-first search). Through our novel unified evaluation framework, we quantify performance across comprehensive metrics: average score, survival time, food collection efficiency, and computational complexity. Our findings reveal that while reinforcement learning methods excel in maximum score achievement (mean: 847.3 ± 12.4), minimax algorithms demonstrate superior consistency (std: 23.1 vs 156.8). Heuristic methods provide optimal computational efficiency with real-time guarantees. These results yield significant insights into algorithm selection trade-offs in constrained gaming environments, with broader implications for AI system design.

Submission Number: 8