Multi-Agent Deep Reinforcement Learning for Efficient Unattended Information Gathering and Monitoring of Autonomous UAM Systems

Abstract

Multi-agent deep reinforcement learning is machine learning in which agents cooperate to achieve a common goal through communication between multiple agents. With this deep reinforcement learning technology, multiple Urban Air Mobility (UAM) can replace the surveillance role of CCTV, which is essential for security and data collection in urban environments. Existing CCTV can provide limited visual information in a fixed location, but building and autonomous CCTV system through UAM can provide flexible and stable visual information according to the location of the surveillance target in real-time. Therefore, this paper proposes a method to build a system where multiple UAMs efficiently perform monitoring services through the CommNet algorithm, which plays the role of inter-agent communication.