StarCraft II AI Agents

This project focuses on the development of sophisticated AI agents for StarCraft II, bridging the gap between simple scripted behaviors and advanced autonomous strategic decision-making.

Core Engineering

Utilizing the python-sc2 library, the agents are built on a modular architecture that separates macro-management (economy, base building) from micro-management (unit control, combat tactics). The 'on_step' decision loop is optimized to handle hundreds of units simultaneously under strict real-time constraints.

Features & Infrastructure

• Protoss Strategy Engine: A specialized engine for Protoss, managing unit production cycles and tactical expansion.
• Headless Docker Deployment: A containerized environment allowing for large-scale automated match testing and data collection on remote servers.
• CI/CD Pipeline: Integrated with GitHub Actions for automated linting and image building, pushing to GHCR for seamless deployment.
• Research Focused: Includes a comprehensive overview of Machine Learning approaches in SC2, analyzing architectures like AlphaStar and various reinforcement learning pipelines.

Results

The project serves as a robust platform for experimenting with game theory and autonomous agents in one of the most complex real-time strategy environments ever created.