An On-Board Control Scheme for Consumer-Level Autonomous Underwater Vehicle in the Intelligent Cyber-Physical Transportation Systems

In recent years, with the expanding scope of the Intelligent Cyber-Physical Transportation Systems (ICTS), Autonomous Underwater Vehicle (AUV) has also become a hot product in this field that consumers are paying attention to. Consumer-level AUVs have a broad consumer market for underwater detection and search and rescue due to their small size, autonomy, flexibility, and long endurance. The formation of multiple AUVs can complete more diverse and complex tasks. However, consumers can only communicate with AUVs when they float to the surface, which poses a tough challenge to AUV's formation control. This paper proposes an AUV On-board Control Scheme (AUV-OCS) based on multi-agent reinforcement learning and cloud computing technology to serve consumer-level AUVs, in which the local perception information preprocessed by each AUV is uploaded to the cloud center for data fusion in real-time and a centralized value function is designed to optimize the global decision. The experimental results show that the proposed AUV-OCS can effectively control the AUV cluster to complete underwater task in a predetermined formation for consumers with an average task success rate of 97.4%, and has a certain collision avoidance ability.