Reinforcement Learning for Digital Twins

Digital TwinsDigital twin (DTs) aim to ongoingly replicate complex systems through data acquisition, simulation, and analysis to monitor, optimize, and/or experiment to achieve systems' goals. Typically, systems adapt and evolve during their lifetimes, which requires updating simulation modelsSimulation models and analysis as new data arrives or conditions change. The dynamics of environments under which DTs commonly operate necessitate using a paradigm that can deal with the uncertainties and unprecedented scenarios that may arise throughout its operation. Reinforcement LearningReinforcement learning (RL) is a learning paradigm that provides tools to do precisely this. It is concerned with sequential decision-making in dynamic, uncertain environments. In this work, we discuss the current and potential role of RL in the context of DTs, motivate its usage through a concrete case studyCase study, and finally discuss the opportunities and challenges.