Cat Mario AI Challenge Project

Overview

This project, inspired by SethBling's MarI/O and Niko's CrAIg, uses NEAT (Neural Evolution of Augmented Topologies) to train an AI for playing "Cat Mario," a challenging Mario-like game. The aim is to develop an AI that learns to navigate through the game's deceptive traps and challenges. For more information, you can also refer to this Medium article.

Dataset and Input-Output Behavior

• Dataset: Real-time gameplay video feeds, converted into frames.
• Input: Game environment information including character, terrain, enemies, and projectiles.
• Output: Simulated keystrokes for the character's movement and actions.

Result

Success Metrics

1. Basic: Completing a training level.
2. Intermediate: Completing an unseen level.
3. Advanced: Improving the fitness function for faster training.
4. Expert: Optimizing the algorithm for quicker convergence.