Dark Side of the Moon: Solving Lunar Lander with Deep Q-Learning
Karina Ponze
Co-Presenters: Individual Presentation
College: The Dorothy and George Hennings College of Science, Mathematics and Technology
Major: Computer Science
Faculty Research Mentor: Israel Curbelo
Abstract:
Title: Dark Side of the Moon: Solving Lunar Lander with Deep Q-LearningAuthor: Karina Ponze, Department of Mathematics, Kean UniversityAbstract:This study uses Deep Q-Networks (DQN), a reinforcement learning neural network-based algorithm, within the Lunar Lander environment of OpenAI Gymnasium. The main objective of this study was to create and train an agent to safely land a spacecraft on rough terrain while simultaneously conserving fuel and reducing crashes. This research tackles the problem of combining continuous state variables, like position, velocity and orientation, with discrete action controls for thruster operations.The methodology incorporates two key reinforcement learning components, a replay buffer and target networks, to stabilize learning and improve adaptability. The agent’s performance was evaluated over 1,400 episodes, cumulative reward, episode duration and success rate. Upon completion of its training, the agent attained an average score of more than 200 points and a success rate of around 47 percent.These findings highlight DQN’s potential in solving complex control tasks in unpredictable environments. This paves the way for applications in areas such as robotics, autonomous vehicles, and energy systems. With algorithm improvements such as Double DQN and Dueling DQN are possible through further research to improve performance.Keywords: Deep Q-Network, Reinforcement Learning, Lunar Lander, OpenAI Gymnasium, Artificial Intelligence