This project is dedicated to the development and comparison of different AI models trained to play the Lunar Lander game. The goal is to successfully land a lunar module on the surface of the moon. The repository includes three distinct approaches: neuroevolution, DQN (Deep Q-Network), and Double DQN.
Neuroevolution is a form of artificial intelligence that uses evolutionary algorithms to generate neural networks, mimicking biological evolution. This model evolves through generations to optimize its landing strategy.
DQN is a reinforcement learning algorithm that combines Q-Learning with deep neural networks. This model learns through trial and error, using a reward system to make better landing decisions over time.
Double DQN is an improvement over the standard DQN algorithm that reduces the overestimation of action values. This model aims to achieve more stable and reliable learning outcomes.
The training progress and results can be monitored through loss graphs which are saved in the assets directory. The images below are limited to 500px in width for consistency in presentation.
