Geometry Dash Double DQN Agent
Summary I created a Double Deep Q-Learning agent and a custom Gymnasium environment for Geometry Dash. Work included observation preprocessing, tailored action space, and reward shaping to make progress on the first level.
Highlights
Double DQN stability improvements over vanilla DQN
Custom Gym-compatible environment
Deterministic evaluation harness
Stack Keras, Gymnasium, Python, NumPy
Links
Repo: https://github.com/bernard0331/CMSC421Project
Project link: https://github.com/bernard0331/CMSC421Project