Optimizing Autonomous Landing Systems with Deep Q-Networks: A Lunar Lander Simulation
Noorul Sama Sahel
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:
The Lunar Lander simulation, part of the OpenAI Gym toolkit, provides a challenging testbed for reinforcement learning algorithms. This project explores the use of Deep Q-Networks (DQN) to train an autonomous agent capable of safely landing a spacecraft while optimizing fuel consumption and stability. The agent learns by interacting with an eight-dimensional state space, selecting from four possible actions to control its descent. A structured reward system encourages smooth landings while penalizing crashes and inefficient fuel use.Through extensive training, the DQN model achieved over 90% success in landing safely, demonstrating improved decision-making and fuel efficiency over time. Performance metrics, such as increasing average rewards and stable Q-value loss, indicate effective learning convergence. These results highlight the potential of reinforcement learning for autonomous control in complex, dynamic environments.By applying DQN to the Lunar Lander challenge, this research showcases how AI can enhance decision-making in robotics, space exploration, and other real-world applications requiring precise, adaptive control.