[Paper Review] Multi-platform Version of StarCraft: Brood War in a Docker Container: Technical Report

Michal Šustr, Jan Malý+1 more|arXiv (Cornell University)|Jan 7, 2018

Artificial Intelligence in Games5 references3 citations

TL;DR

This paper presents a Dockerized, multi-platform solution for running StarCraft: Brood War with integrated AI development tools, enabling cross-platform deployment, reproducible testing, and scalable training of AI bots. By containerizing the game via Wine and pre-installing BWAPI and Java support, the system allows researchers to easily run, test, and scale AI agents across diverse environments with configurable resources and headless or GUI modes.

ABSTRACT

We present a dockerized version of a real-time strategy game StarCraft: Brood War, commonly used as a domain for AI research, with a pre-installed collection of AI developement tools supporting all the major types of StarCraft bots. This provides a convenient way to deploy StarCraft AIs on numerous hosts at once and across multiple platforms despite limited OS support of StarCraft. In this technical report, we describe the design of our Docker images and present a few use cases.

Motivation & Objective

• Enable cross-platform deployment of StarCraft: Brood War AI agents despite native OS limitations (Windows/Mac only).
• Provide a self-contained, reproducible environment for testing and training AI bots with consistent system resources.
• Facilitate large-scale parallel execution of StarCraft games for machine learning and benchmarking.
• Simplify integration of diverse bot types (C++, Java, etc.) by pre-installing required tools and APIs like BWAPI.
• Support both headless and GUI modes for efficient training and debugging of AI agents.

Proposed method

• Containerize StarCraft: Brood War using Wine to run on Unix-like systems, enabling cross-platform compatibility.
• Extend the Wine base image to include StarCraft: Brood War, BWAPI, and necessary game data and maps.
• Build dedicated Docker images with pre-installed Java environments to support Java-based AI bots.
• Implement configurable resource allocation per container to ensure reproducibility and performance isolation.
• Use Docker volumes to share game data, logs, and cache (e.g., BWTA2 map analysis) between containers and the host.
• Support both headless (GUI disabled) and headful (GUI enabled via VNC) modes for flexible deployment and debugging.

Experimental results

Research questions

• RQ1 How can a consistent, multi-platform StarCraft: Brood War environment be created to support AI research across diverse systems?
• RQ2 What is the impact of containerized resource isolation on the reproducibility and reliability of AI bot performance testing?
• RQ3 Can Docker-based deployment simplify the integration and execution of diverse AI bot types (e.g., C++ and Java) in a unified environment?
• RQ4 How effective is the containerized setup for large-scale parallel training of AI agents in self-play or benchmarking scenarios?
• RQ5 What role can Docker play in reducing deployment complexity and licensing barriers in AI research using proprietary games like StarCraft: Brood War?

Key findings

• The Dockerized environment successfully enables StarCraft: Brood War to run on any system supporting Docker, overcoming native OS limitations.
• Pre-installed tools such as BWAPI and Java environments allow immediate use of diverse bot types without manual configuration.
• Resource allocation per container ensures consistent and reproducible results across different hardware setups.
• Headless mode supports efficient large-scale training by eliminating GUI overhead, while VNC access enables debugging in headful mode.
• The use of shared volumes for logs, caches (e.g., BWTA2), and game data enables efficient data sharing and caching across containers.
• The system supports automated bot retrieval and execution via SSCAIT tournament API, streamlining benchmarking and testing workflows.