Machine Learning Enhanced Autonomous ADCS for an IoT CubeSat

Abstract This paper discusses the design and development ofa hardware prototype of an Autonomous Attitude Determination and Control Subsystem (ADCS) for an Internet of Space Things (IoST) CubeSat with Machine Learning (ML) integrated into the system. It is a novel design to the existing ADCS by integrating ML into it and making the whole system autonomous thereby improving reliability and efficiency. The prototype developed is based on the design and structure of a one-unit (1U) CubeSat which are widely used in the IoST domain. A CubeSat encompasses several subsystems, out of them ADCS holds significant importance. The ADCS must be designed within stringent mass, size and energy constraints. Additionally, CubeSat typically has limited sensing, computing, and communication capabilities, highlighting the necessity for autonomous and computationally efficient algorithms. In such constrained environments, developing an effective attitude control system presents a significant challenge. The proposed design discussed in the paper upgrades the existing ADCS to a better efficient and reliable one with optimum energy consumption and less failure rate with the integration of ML technologies and enabling it to be autonomous. The selection of hardware and algorithms takes into account system requirements, limitations, past experience, and both theoretical and practical considerations. The paper thoroughly explores the design methodologies, hardware components, and algorithms utilized in the experimental hardware prototype.