The paper describes a system in which a sensory-equipped, two-degrees-of-freedom gripper and its controller form an autonomous unit. The aim was to create a system capable of local decision-making, working within the hierarchically organized robotic cell. The microprocessor-controlled gripper system makes decisions based on the sensory information and on partial knowledge of both the task and the status of the cell. Specific reactions of the gripper to sensory inputs can be consistently repeated in a whole range of different tasks. Moreover, the system processes the majority of the sensory signals, leaving to the cell supervisor only the control of the flow of the main task. Elementary reactions are determined by responses of the servo loops, while other, more sophisticated reactions may be acquired by learning. It is shown how the capabilities of a relatively simple multiple-loop analogue controller benefit from this feature.