HTN guided game tree search for adaptive CGF commander behavior modeling

摘要

Employing Hierarchical Task Network (HTN) for commander behavior modeling usually requires a rich set of knowledge in order to be responsive to various situations during the combat. This paper presents an HTN planning based game tree search to enhance the commander agent's deliberation capability. Given a general HTN structure, our approach evaluates different decomposition branches through look-ahead reasoning, and produces appropriate decisions. Compared with using HTN alone, the HTN guided tree search can online explore tasks' adaptabilities to different situations, thus reducing the impact of HTN knowledge insufficiency. We apply this approach to an infantry combat simulation, where the commander needs to guide three platoons to clear enemies in specific areas. Results show that it can effectively find the best strategy from HTN encoded alternatives.