gym-starcraft

Why we included this project

For researchers who want to train reinforcement learning agents in a real-time strategy game, gym-starcraft wraps StarCraft into the standard OpenAI Gym interface, so any RL algorithm that works with Gym can be plugged in without custom glue code. It builds on Facebook's TorchCraft to bridge the game engine and the Torch stack, which makes it a natural fit for teams already living in that ecosystem. The repo ships a working random-agent example and step-by-step setup instructions for launching a StarCraft server over the network, so you can get moving faster than you would building your own bridge from scratch. That said, it targets the classic Brood War client through BWAPI and has seen no commits since 2017, so budget for maintenance if you want to base anything serious on it.