Modeling and control architecture for the competitive networked robot system based on POMDP

Abstract

As the competitive networked robot system has the characteristics of strong interaction and high real-time request, the present control methods which are mostly used for the collaborative networked robot system may not be directly applied to the competitive one. Thus the hierarchical control architecture for the competitive networked robot system is proposed in this paper in order to adapt to its two characteristics above. To deal with the system observation uncertainty caused by noise and time delay, and the system action uncertainty caused by the opponent in the meantime, the control architecture based on Partially Observable Markov Decision Processes (POMDP) has been adopted by the executive layer to select the action of the maximum expected reward, thus fulfilling the intention of the strategy layer effectively. In addition, the introduction of the executive layer has successfully freed the strategic layer from the tasks that should have been completed by its bottom layer alone, thus enabling the strategic layer to focus more on its strategic design. In this paper, the networked robot system named Tele-LightSaber with a high degree of confrontation is designed and implemented, and the experiment results show the validity and efficiency of the proposed method on TLS platform.