As part of an ongoing partnership with AI research firm DeepMind, Blizzard has released a public API for StarCraft II geared at furthering AI research.
The release makes it possible for developers and researchers alike to test and observe AI agents as they go about any number of complex tasks that make up a StarCraft II match.
Both parties have high hopes that the open source project will play a part in accelerating AI research. To that end, the pair has made a number of resources available to the public.
Those include a scripted AI API, image-based AI API, documentation, sample code, sample bots, offline AI vs. AI support, a fully-featured Linux package of the API, and an open source version of DeepMind's PySC2 toolset.
While this isn't the first time AI researchers have turned to video games as training grounds for AI agents, DeepMind notes that the sheer complexity of a real-time strategy game like StarCraft II makes it an ideal environment for AI research.
DeepMind notes that a StarCraft player has more than 300 basic actions at their disposal, compared to the 10 or so offered by Atari games at any given time. The recently released StarCraft II API, coupled with DeepMind's PySC2 environment wrapper toolset, means that researchers and developers alike can create and train AI to take on the level of challenge posed by Blizzard's RTS game.
More information on how to get started with the StarCraft II API, as well as various GitHub directories for different elements of the project, can be found over on Blizzard’s blog. Additionally, information on DeepMind's open source PySC2 AI toolset can be found over on its own blog.