Introduction

This project showcases an AI that autonomously plays Pokémon Platinum by analyzing in-game visuals to decide how to navigate the world. The AI operates by capturing two images: one before and one after a move, then subtracting them to detect changes. A Convolutional Neural Network (CNN) classifies these changes to identify movement or interactions.

The AI dynamically maps out the game environment, placing walls on the map where it detects obstacles and marking spaces where no wall is detected. To handle cutscenes or battles, the AI opens the game menu to check if it is possible. Additionally, the AI can handle warps in doorways and knows which maps connect to each other, allowing it to remember areas that are not currently visible.

Technology

This project utilizes several key technologies to build and train the AI model effectively:

• Python: The primary programming language used for building the environment and implementing the AI logic.
• TensorFlow: A deep learning framework used to implement and train the CNN to detect movement.
• Pillow: A library used for taking screenshots.

Project Repository:

For more details, visit the project repository on GitHub.