Starcraft 2 is a very complicated strategic video game, but it's number of possible play combinations exceeds games like Chess by millions of times. There is no single way to play and no single good strategy to win a game. This is why, in such a difficult game we would want to create an agent that is able to exceed a human player and win. A simple bot would not be enough, because, even taking into account a limited amount of strategies, the code base would become immense just in a couple of play moves, so a RL agent would be more efficient and would in the end give us better results for a variety of situations, taking into account the adaptability to new situations.
- TechnologiesStable-Baselines3, Pytorch, Python, Wandb, Numpy, OpenCV
- GithubStarcraft2 Reinforcement Learning
Used resources
