Unified learning to enhance adaptive behavior of simulation objects

Abstract

Modeling and simulation are methods of validating new systems that are risky to be directly deployed in the real world. During the simulation, the simulation environment continuously changes and simulation objects correspondingly behave according to the changing situations. In general, modeling the behavior for all possible situations is extremely difficult when the rationale is unknown. Therefore, in order to adapt to the changing situation, it is important to recognize the rationale behind the behaviors of the simulation object. However, in many cases, even though the rationale is unknown or difficult to recognize, the simulation requires reasonable behaviors such as a commander’s decision in a war game simulation and a driver’s behavior in rush hours. In this study, we propose a new approach to determine the behavior of simulation objects under changing situations. The proposal is a unified learning approach that integrates two methods, data-driven and knowledge-driven approaches, which allow simulation objects to learn behavioral knowledge from experience as well as from domain experts performing the simulation and reuse verified knowledge. By combining both approaches, we supplement the shortcomings of one method with the strengths of the other. To verify our method, we apply the proposed approach to a military training simulation.