Abstract
Use Deep Q-Networks (DQN) to train an autonomous agent to play a custom-built Chrome Dinosaur game in Python. A CNN-based model processes gameplay frames and obstacle information to perform real-time decision-making and obstacle avoidance.
Skills
Awards
Link
Contents
This is a 3 people group project! :)
This project explores the application of Deep Q-Networks (DQN) in a custom-built Chrome Dinosaur game environment developed in Python. Instead of using the browser version directly, the game environment was recreated to provide full control over obstacle spawning, game speed, reward design, and state representation. A CNN-based reinforcement learning model was trained using consecutive gameplay frames and obstacle distance information to enable real-time decision-making and autonomous obstacle avoidance. Through experiments with replay buffers, epsilon decay, and reward engineering, the project investigates how reinforcement learning agents learn stable behaviors in dynamic interactive environments.
