SCC: Deep Reinforcement Learning Agent plays StarCraft II at competitive human level

StarCraft is considered to be one of the most challenging RTS games for human and AI research. StarCraft Commander(SCC) reaches competitive human level playing StarCraft II full game, by using neural networks trained from supervised learning and deep reinforcement learning algorithms, with limited computation resources. During the demonstration, users will be able to play with a diverse set of agents that exhibit different skill levels, plays styles and strategies. We will also demonstrate various visualizations about how the agent learns, thinks and acts. We aim to provide insights into the learning agent and technology behind it during this interactive demonstration.