A two-step machining and active learning approach for right-first-time robotic countersinking through in-process error compensation and prediction of depth of cuts.