A Five-Dimensional Digital Twin Architecture for Real-Time and Non-Real-Time Safety Assurance of Autonomous Systems

The rapid advancement of autonomous systems has revolutionized various industries, yet ensuring their safety remains challenging. This paper explores the potential of digital twins —virtual replicas of physical entities— to enhance the safety assurance of autonomous systems. Digital twins offer a dynamic solution for real-time safety validation by enabling continuous monitoring, testing, and optimisation. The paper reviews current applications of digital twins in autonomous driving, collaborative robotics, and aerospace, highlighting their effectiveness in simulating operational scenarios and optimizing system performance. Despite their promise, several challenges persist to using digital twins to assure the safety of autonomous systems at runtime, including data integration, real-time processing, scalability, and model selection. To address these, we propose a novel five-dimensional digital twin architecture encompassing data collection, preprocessing, integration, visualization, and real-time and non-real-time components. We discuss various techniques for leveraging digital twins in safety assurance, such as virtual sensing, data comparison, scenario testing, and continuous monitoring. A case study on an autonomous mobile robot in a warehouse setting demonstrates the practical application of our proposed architecture, showing improvements in safety and efficiency.