Project Abstract

With artificial intelligence currently being at the forefront of research in computer science, many projects are exploring new approaches to enhance AI capabilities, ranging from deep learning algorithms to reinforcement learning frameworks. The drive to develop AI systems capable of human-like reasoning and decision-making processes has become a central theme which is driving collaboration across disciplines and breakthroughs in areas such as natural language processing, computer vision, and robotics.Joining the thriving field of AI, this project explores playing the game of Tetris by utilising deep reinforcement learning techniques to help train an AI agent that is capable of setting competitive scores while surviving for extended periods of time. The study explores strategies arising from advanced features such as piece holding and predictable piece queue randomisation, which have been introduced in more modern editions of Tetris. By harnessing deep Q-learning within a custom made environment, the study’s main aim was to optimize the agent’s decision-making process through a combination of techniques such as epsilon-greedy policy, experience replay memory, and a linear fully connected network. The project presents the effectiveness of its approach in guiding the agent towards known Tetris strategies that are commonly employed by human players with results that discover the potential pitfalls of risky Tetris strategies and how an important balance must be struck when attempting to maximise the agents performance. Results additionally show how the agent is able to utilise these additional game mechanics effectively to reproduce advanced strategies that go beyond the core game mechanics. By investigating neural network solutions and alternative feature representation for control issues, this project may offer to assist researchers in exploring fields such as autonomous vehicles and any further study into the subject of Tetris AI.

Keywords: Artificial Intelligence, Reinforcement Learning, Deep Neural Networks

 Conference Details

Session: Poster Session A at Poster Stand 25

Location: Sir Stanley Clarke Auditorium at Tuesday 7th 13:30 – 17:00

Markers: Eike Neumann, Fernando Maestre Avila

Course: MSci Computer Science, 3rd Year

Future Plans: I’m continuing studies