Reinforcement Learning for the Beginning of Starcraft II Game

Starcraft II is a popular real-time strategy game that is welcomed by many young people. This game is really complicated to play due to the long game timeline, various society/buildings/units, a large number of actions or selections, different constraints (e.g., population limit), and the partially observed environments. In this project, we plan to develop a reinforcement learning model for the beginning of Starcraft II game, instead of the full-length game. The beginning of the game is essential for the further economy, population increase, and technology development. Our project is based on the SC2LE (StarCraft II Learning Environment) platform. We build a feasible pipeline for training reinforcement learning models and design random, scripted, and our actor-critic based agents. Experiments show that our actor-critic based agents can learn valuable knowledge in this task. The video is available at https://drive.google.com/file/d/11S6t3rNKjM1CJEkFSN1lkZiyLTdqHhzQ/view?usp=sharing