AlphaStar: an integrated application of reinforcement learning algorithms
Yulong Zhang, Li Chen, Xingxing Liang, Jing Yang, Yang Ding, Yanghe Feng · 2022
Deep Reinforcement Learning (DRL) is poised to revolutionize the field of AI and represents a step towards general intelligence. Currently, AlphaStar achieved the Grandmaster level in StarCraft gaming, which is a remarkable breakthrough in Real-Time Strategy (RTS) gaming. The paper discusses the general framework of RTS gaming agents, and presents the method innovation from baseline RL algorithms to the AlphaStar. We begin with the basic idea and taxonomy of DRL, then progress to the technical framework of AlphaStar from the perspective of state space design, action space design, policy and value net framework and multi-agent training methods, presenting key issue and feasible method in full-length scale of RTS gaming. Finally, we draw a brief conclusion.