Research On CGF Behavior Model Based On Behavior Tree

In the combat simulation system,if the lack of effective modeling of human behavior will directly affect the effectiveness of the target system and the credibility of simulation results,therefore,to build computer-generated force model is an important work of simulation system construction,The simulation of human warfare behavior to build a credible virtual opponents,friendly,and intelligent models such as civilians.In the CGF modeling,the most central question is how to enhance the CGF behavioral fit to accurately reflect the objective behavior of the simulation object,that is,CGF behavior modeling.In the traditional CGF modeling,the description and control of the behavioral execution logic mainly use Finite State Machine(FSM).FSM is widely used and developed in all kinds of simulation platforms,such as ModSAF.However,with the increase of modeling complexity,its modularity is low,and the shortcomings of "one-step control" and difficult to integrate development are gradually revealed.Behavior Tree(BT)is a behavioral description technology developed in recent years in the field of Game AI.Compared with FSM,BT has a high degree of modularity,"two-step control",hierarchical description structure And other advantages in the commercial game market to occupy an increasingly high share.This paper attempts to introduce a behavioral tree into the combat simulation system,seeking to achieve a more effective description of CGF behavior.Similar to the FSM,the basic behavior tree is essentially a static behavioral organization.It can not be used to construct the CGF model dynamically by using the behavior tree,and it is difficult to simulate the process of human behavior model through experience accumulation and learning..In order to solve this problem,this paper studies the CGF behavior modeling framework,the dynamic combination of behavior tree and the structure optimization based on the analysis of the structure and control principles of the basic behavior tree.In view of the first problem,this paper constructs the CGF behavior modeling framework,which is the basic structure of CGF behavior organization.By designing reasonable function module,CGF can make the environment,The behavioral tree technology is used to manage the tactics.Then the CGF behavior control technology is discussed.For the second problem,this paper proposes a method of dynamically constructing behavior tree tactics strategy based on rule reasoning.This method realizes the dynamic combination of behavior tree based on production rules by designing reasonable knowledge architecture.In this paper,the knowledge representation method based on production rules,the rule-based tactical behavior planning mechanism and the behavior tree dynamic construction method based on the planning results are expounded in detail.For the third problem,this paper adopts the reinforcement learning method to conduct on-line learning and dynamic optimization of tactics strategy based on behavior tree.By using the consistency between the behavior tree and the hierarchical reinforcement learning,the tactical strategy can be optimized and converged step by step in order to make the CGF learn at the tactical strategy level,not only at the reactive action level.In this paper,we introduce reinforcement learning function into the behavior tree by introducing the Q-Learning selection node to replace the original selection node,and choose the tactical strategy according to the stable Q value information obtained after learning.In this paper,a simulation experiment is designed and implemented based on ORTS game platform.Soar reasoning engine is used to construct the planning framework.The online optimization of the behavior tree is preliminarily verified by the confrontation with the script AI provided by the platform.