This contribution aims to analyze medical interventions scheduled and executed using image-guided robots. Such interventions, which can be surgical or restricted drug delivery, involve minimally invasive, precise, and harmless therapies. The accuracy of robotic positioning is trained via uncertainty and complexity reductions. This could be achieved by matching real and virtual procedures involving physical and virtual phantoms of the relevant part of the living tissues. Such tailored training includes personalized, patient and interventional tool features. The training results enable a real (with patient) intervention controlled by staff and a possible matched autonomous intervention under staff supervision. The paper includes regards on the control and monitoring of image-guided procedures, and on augmented matched digital twins for real interventions. The various topics covered in this narrative review, are supported by examples from the literature.