Back to projects
May 08, 2025
2 min read

AI to Play a 3D Two Player Strategy Game (Columns)

Developed an AI using NEAT to play the strategic stacking game Columns, evolving neural networks through 1.12M simulated matches.
See Demo See Repository

Overview

This project showcases an AI developed to play Columns (a two player strategic stacking game), using a NEAT (NeuroEvolution of Augmenting Topologies) algorithm to evolve neural networks. The AI competes against itself, improving over generations through simulated natural selection.

AI Training with NEAT

  • Configurable population size, generations, incentives and topologies.
  • High-performing networks evolve to improve gameplay.

Development Progress

  • Training for 150 genomes over 100 generations (1.12 million games) resulting in a functional AI capable of reasonable gameplay.
  • Command-line interface with text-based game outputs.

Planned Enhancements

  • 3D Visualization: Create an interactive game display.
  • Efficient Storage & Review: Reduce game file size, enable multi-game analysis with move trees.
  • Genome Uploads: Save trained AI models for user interaction.
  • Evaluation Metrics: Implement position evaluation similar to chess engines (+1/-1 scoring).
  • Learning Optimization: Refine incentives and topology for strategic improvement.
Prev
Game Development Projects
Next
Task Tracker CLI