Game Tree Search Based on Nondeterministic Action Scripts in Real-Time Strategy Games

摘要: Significant progress has been made in recent years toward stronger real-time strategy (RTS) game playing agents. Some of the latest approaches have focused on enhancing standard tree search techniques with a smart sampling space, or directly reducing this space. However, experiments thus far only performed using small scenarios. We provide experimental results performance these agents increasingly larger Our main contribution is Puppet Search, new adversarial framework that reduces space by scripts can expose choice points to look-ahead procedure. Selecting combination script and decisions for its represents an abstract move be applied next. Such moves executed actual game, representation state, which used algorithm. tested Search μRTS, RTS popular within research community, allowing us compare our algorithm against state-of-the-art published last few years. show similar other scripted based smaller scenarios, while outperforming them ones.