On the Use of Multi-agent Reinforcement Learning in Cyber-physical and Internet of Thing Systems.

In this paper, we provide a review of cyber-physical systems (CPSs) and explore the applications of Multi-Agent Systems (MAS), Multi-Agent Reinforcement Learning (MARL), and Multi-Agent Credit Assignment Problem (MCA) in CPSs. Our primary focus is on mapping specific domains, including job scheduling, energy management, and smart transport systems, to MAS and applying MARL and MCA techniques to solve the problems. To evaluate the effectiveness of our proposed method, we applied it to the job scheduling problem, using two parameters, CPU and bandwidth, and tested its performance for four different tasks: Face Detection and Window Blind Control (FDWC), Finger Touch and Gate Control (FTGC), Weather Check and Thermostat Control (WCTC), and Temperature Check and Start Fan (TCSF). The results indicate that prioritizing tasks significantly improves the performance of the proposed method. We conclude that MAS, MARL, and MCA are powerful tools for solving problems in CPSs and IoT. Mapping these problems to MAS can help overcome the challenges associated with CPSs and IoT, and improve system performance.