Meta-reasoning for Cognitive Digital Twins: High-Level Architecture and Roadmap

A digital twin is a virtual representation of a physical system, which can be used for prediction and control. This representation can include predictive models and simulations, which are continuously updated with physical system data. Autonomous control using a digital twin can be considered as a cognitive loop, involving interpretation of sensor data (perception), integration of the data into predictive models (which can involve reasoning) and the planning and execution of actions on the physical system. In particular, meta-reasoning can be used to monitor and control the algorithms and models of the digital twin. This is important for autonomous problem detection and recovery, as well as for explanation and transparency.Our approach to cognitive digital twin systems is first to specify a high-level architecture, grounded in cognitive systems research (as is often done in robotics). We then provide a background and survey of meta-reasoning and include it as an additional layer in the architecture. Finally, we outline a roadmap for gradually increasing the autonomy and resilience of a digital twin system by iterative development in a real-world environment.