Project Abstract

The motivation behind this project stems from the growing interest in applying machine learning to an AI to make real-time decisions to solve problems in a dynamic environment like Tetris, a classic game known for its dynamic and unpredictable setting. With machine learning methods becoming more accessible and powerful, comparing their effectiveness to each other can offer valuable insights into their strengths and limitations for such decision-making tasks. This research aims to compare and evaluate the performance of two different machine learning approaches – Deep Q-Learning and Genetic Algorithms – in training an AI agent to play Tetris. The focus is on identifying which method better adapts to the game’s increasing complexity and delivers more consistent performance. The design involved implementing the game of Tetris and then implementing both algorithms within the same Tetris environment, running multiple training iterations, and recording metrics such as highest score, average lines cleared, and learning speed. The findings show that while the Deep Q-Learning model adapted more quickly to early-game scenarios, the Genetic Algorithm exhibited better long-term performance and stability over time. Ultimately, the project produced two functional AI agents and a comparative analysis framework. This work aims to contribute a better understanding of how different machine learning strategies handle non-deterministic, time-sensitive decision-making tasks, offering insights for future AI development in games and more.

Keywords: Machine Learning and AI, Deep Q-Learning, Genetic Algorithm

 Conference Details

Session: A

Location: Sir Stanley Clarke Auditorium at 11:00 13:00

Markers: Eike Neumann, Bertie Muller

Course: BSc Computer Science 3yr FT

Future Plans: I’m looking for work