Research On Agent Behavior Modeling In RoboCup Rescue

This thesis focuses on the development of a noval approach for modelling the behaviors of agents in a typical multi-agent System namely the Robocup Rescue Agent Simulation. Meanwhile, this model is applied to regulate and promoe the development of agents so as to help agents to cope with the challenges in the disaster environment in a more effective and efficient way.The Robocup's rescue agents'behaviors being analyzed and definied in the beginning, this paper divides agents'behaviors into two categories:decision-making behavior and action behavior, in which the former decides the latter. On this basis, a task oriented hierarchial decision model is put forwad in the present paper. There are three layers:the underlying layer, the tactial level and the strategic layer in this model, in which the underlying layer includes the action module and the processiong module, the tactical layed is in charge of task selection and the strategic layed is mainly responsible for task allocation.There are three main characteristics in the task oriented hierarchial decision model:first and formost, it is more of a reader-friendly model because of iys task oriented essence. Besides, as a multilayer model, it can avoid the intrisic defects of single models wherever possible. Lastly, thanks to the task selection and task allocation in its decison making, this model can balance between both local information and global information, which contributes to more adaptable and rational agents' bahavior.In light of the task oriented hierarchical decison model, this paper has designed and developed both firebrigade and ambulance team. Moreover, a in-depth research on the most complex and a rare involved agents' behavior namely multi-agent coordination in RMASBench, derived from the RoboCuo rescue system is carried out in this study. The interwoven coordination of RMSBench agents is exploited through the same model.A series of experiments are carried out correspondingly and all experimental results show a ood overall performance of the agents developed whose behavior are proved to be efficient and robust.