During the workshop Embodied AI and Edge Computing for Intelligent Robots at IROS I presented several our advances in controlling manufacturing robots to reach autonomy and adapt to various manufacturing scenarios. One of the important technologies to reach that is metalearning, on which we work actively. Using interactive optimization (Bayesian optimization, reinforcement learning), and adapting models or controllers in the cloud enables even better adaptation capabilities. I also highlighted an upcoming publication (Multifidelity Guided BO with DIgital Twins), which also demonstrates how edge and cloud computing increase autonomy and adaptability for robotic and manufacturing systems and components.