Improvements and Challenges in StarCraft II Macro-Management A Study on the MSC Dataset

Abstract

Macro-management is a crucial aspect of real-time strategy (RTS) games like StarCraft II, which involves high-level decision-making processes such as resource allocation, unit production, and technology development. The MSC dataset, as presented in the original paper, provided an initial platform for researchers to investigate macro-management tasks using deep learning models. However, there are limitations to the dataset and existing baseline models that call for improvement. In this paper, we discuss the challenges and opportunities in enhancing macro-management research in StarCraft II. We propose improvements to the dataset by incorporating new features, addressing data imbalance, and updating preprocessing techniques. Furthermore, we review recent research and state-of-the-art methods in RTS games, such as attention mechanisms, graph neural networks, and reinforcement learning, that could be applied to improve existing tasks or introduce new research directions. We also present our experimental results, highlighting the effectiveness of the proposed improvements and novel approaches. Our goal is to inspire the research community to explore advanced AI techniques and strategies in macro-management and contribute to the development of more capable AI agents in complex RTS games.