HISTORY-BASED SINGLE BELIEF STATE GENERATION FOR PARTIALLY OBSERVABLE REAL-TIME STRATEGY GAMES

Abstract

Researches of AI planning in Real-Time Strategy (RTS) games have been widely applied to human behavior modeling and war simulation. Due to the fog-of-war, planning in RTS games need to be implemented under partially observable environment, which poses a big challenge for researchers. This paper focuses on extending Hierarchical Task Network (HTN) Planning in partially observable environment, and proposes a partially observable adversarial hierarchical task network planning with repairing algorithm named PO-AHTNR. By adding sensing action into HTN domain knowledge, a reconnaissance strategy and a history-based single belief state generation method are presented to obtain the best action. In order to verify the proposed algorithm, an empirical study based on µRTS game is carried out, and the performance of modified algorithm is compared to that of AHTNR and other state-of-the-art search algorithms developed for RTS games.