Flexible and reactive control of interaction between humans and robots will allow a closer cooperation in all service and industrial tasks, that require the adaptability skills of humans to be merged with the high performance of robots in terms of precision, speed and payload [2]. For this reason, recent research strives for intuitive human-robot cooperation, avoiding explicit clarification dialogue and commands. We believe that sensor-based methods, such as visual servoing, provide better solutions, for intuitive HRI, than planning techniques requiring a priori models of the environment and agents. To our knowledge, few works [2-4] have explicitly dealt with HRI using sensor-based control approaches. Our objective is similar: we aim at enabling intuitive interaction between human and robot, in the context of an industrial scenario, where the two must collaborate to realize a task. The robot must be able to infer the human intentions during the task, using only sensed data. The objective of this work is to enable a robot to aid a human operator in a screwing operation.