Deep Q-Learning Aircraft Landing Control
This project explores how different hyperparameters affect deep q-learning performance through systematic experimentation. Using a custom gymnasium environment simulating aircraft descent, I investigated the complex relationships between batch size, learning rate, discount factor, and exploration strategies.
The neural network architecture consisted of an input layer, 3 hidden layers (256 neurons each), and an output layer, utilizing ReLu activation functions. This setup enabled the agent to learn control policies for managing descent rate, angle, and positioning while maintaining stability throughout the landing sequence.
Key Achievements
Custom Environment
Developed a gymnasium environment for aircraft descent simulation with physics-based reward shaping
Parameter Optimization
Conducted comprehensive studies across batch sizes, learning rates, discount factors, and exploration rates
Performance Improvement
Achieved 51% above average performance (397.9 vs 263.5) through optimal parameter selection
Key Learnings
Deep Q-Learning Implementation
Gained hands-on experience in implementing and optimizing deep q-learning algorithms
Reward Engineering
Mastered the art of reward shaping for complex control tasks, learning through multiple iterations
Parameter Sensitivity
Understood the delicate balance between learning stability and performance through hyperparameter tuning