Research And Application Of Semi-Autonomous Combat Agent Model

In the research on the military combat simulation, in order to make the virtual environment is more like the real battlefield and the object which behaves in the simulation plays an more living combatant, the concept of Semi-Autonomous Combat Agent is propsed. It is an intelligent agent who can complete its assignment in the dynamic uncertain battlefields, it has high autonomy but partly controlled by other agents or constrained by other factors that is called incomplete autonomy. Combining the autonomy and restriction into entia, Semi-Autonomous Comabt Agent can react with the agents in the environment proactively or passively. Semi-Autonomous Combat Agent is autonomous, proactive, social and intelligent. But it must be obey the order sent by the higher level agent such as the commander in the battlefield. Semi-Autonomous Combat Agent play an living combatant in the battlefield simulation. Faced on the weapon evaluation applying the combat simulation, combining the characters of combatant/weapon and the organization of commandder, a Semi-Autonomous Combat Agent based combat model is proposed. And an battlefield combat simulation system is constructed on that model.We begin by defining the concept and designing the construction of Semi-Autonomous Combat Agent. The communication model, learning model and bounded rational decision model of Semi-Autonomous Combat Agent is the main research part of this paper. We expand some traditional method or outline some new arithmetic in these researches. First, we propose Semi-Autonomous Combat Agent Communication Model (SCACM) which is in charge of the communication between combatant agents who play in the battlefield. The SCACM use the blackboard model and Agent Communication Language (ACL) model as the basic communication mechanism. As a core particular method, proactive information exchange (PIX) approach is propsed. The PIX adapt to communicate between agents belong to the same combat team agents in battlefield. The major benefits of PIX demonstrated in this paper are reduction of information exchanged occupying on the bandwidth and reduction of redundancy. Relying on the effective communication, the task assignment model for cooperation is based on the set covered theory. An cooperation request arithmetic used to resolve the problem of coordination between two agents in one team. All the methods talked above have been examined in real instance.Second, we emphases on the agent learning method. The character of learning of agent make it can improve his ability and experience to resolve problem in the process of completing his task. The profit-sharing reinforcement learning approach based on Semi-Autonomous Combat Agent is proposed in this paper can avoid tow traditional problem which occur in the process of agent learning, they are Perceptual Aliasing and Concurrent learning. The role allocation and cooperation among combatant agents are used in this learning approach for therestriction of Semi-Autonomous Combat Agent. Through knowledge sharing and higher agents making orders, the learning experience can be shared among the appointed agents who is often belong to one team. This approach can speedly improve the abilities of the agents who share the experience.Third, the decision model is described. A lot of uncertainties and indeterminism in the battlefield environment make the state transfer is not satisfied with the condition of Markov Decision Process (MDP). The bounded rational Semi-Autonomous Combat Agent who make decision based on the expectation utility theory is proposed, that quantify the criterion of making decision. The bounded rational agent decision should consider the cost of computation and consumption of other uncertain factors. This quantity decision arithmetic can avoid the subjective notion arose by the traditional agent reasoning decision method. This expectation utility decision mehtod is more suit of agent behavior selection in the battlefield simulation.At the end, a tank combat efficiency evaluation system is introduced. The Semi-Autonomous Combat Agent combat model is applied in the tank combat simulation subsystem which is part of the tank evaluation system. This combat simulation subsystem is construct on the HLA. We build a tank combat federation in charge of the simulation running, include white federate, red tank federate, blue tank federate and assistant weapons federate. Applying all the function model which introduced in this paper, the Semi-Autonomous Combat Tank Agent is the main reserch object. And using the simple "react" Semi-Autonomous Combat Agent to model mass assistant weapons.Semi-Autonomous Combat Agent have "autonomy" and "restriction", that is more powerful and flexible in describing the entity which is act in the real social environment than the traditional intelligent agent. Besides the military simulation, Semi-Autonomous Agent can be used in ITS (Intelligent Transaction System) etc. Semi-Autonomous Agent used in the research for prediction, planning, analysis of the system can play well, too.