Our objective is to extend the assistance function- alities of the cable-driven parallel robot Marionet-Assist, designed principally for transfer operation, by allowing it to grasp usual objects (knives, box of medicines, phone, ***) by using visual servoing. Our crane robot has a configuration that provides three translational d.o.f., and a camera was added to its end-effector. In order to compute the translational velocity sent to the robot controller, the area and center of gravity in the image of the object to be grasped are used. Experimental results are presented. They show the robustness of our scheme with respect to modeling errors and an excellent positioning accuracy allowing grasping.