Cooperative capture by multi-agent using reinforcement learning application for security patrol systems

Aim of this study is to create a security patrol system which is a cooperative capturing system by using multi-agent in the building. A host computer deploys autonomous robots as agents and find the best strategy to enclose an intruder. From the view point of the pursuit problem by multi-agent, reinforcement learning theory is one of choices to find way how to enclose an intruder. In order to apply reinforcement learning theory to security patrol systems, this study introduces how to discretize patrol areas. Some RFID tags are embedded in the floor and each autonomous robot can know the location where it is by sensing RFID tags, then sends locational information to the host computer. The host computer calculates positioning of the autonomous robots based on received locational data through wireless network. We make a prototype of patrol system and show how it works in this paper.