This paper describes the sensing and control elements of a system for automated robotic edge deburring. The deburring path, automatically generated bp a task planner, is corrected on-line by an active end effector with the objective of controlling the chamfer depth. The sensing system combines the information from force and vision sensors during deburring to provide an improved depth measurement. The vision sensor is then used to verify the deburring performance during an inspection pass. The control system incorporates a new form of adaptive Generalized Predictive Control (GPC) combined with learning control, termed GPC with Learning (GPCL). The system is tested through computer simulations and deburring experiments. The experiments were performed on steel parts with one-dimensional (1-D) and two-dimensional (2-D) edges. For the 1-D edges the depth's standard deviation measured on-line was 0.015 mm with nonadaptive GPC, 0.009 mm with adaptive GPC, and 0.006 mm with adaptive GPCL. With adaptive GPCL and the 2-D edge the deviation was 0.017 mm. This was confirmed by the inspection pass measurements which reported a mean of 0.39 mm and a deviation of 0.019 mm.