Many of today´s industrial robots are still programmed to follow trajectories. The robot is guided by a human operator to the desired fixed application locations. These motions are recorded and are later edited, within the robotic language residing in the robot controller, and played back, for the robot to be able to repetitively perform its task. This methodology is enough to move objects between fixed locations. As the robot is trained to move within fixed positions, all manipulator parts, as well as all the objects which will be manipulated need to be in well defined positions, otherwise another program is needed. Another methodology would be teleoperation, where the robotic system`s movements are executed in manual mode, having the operator working in a master-slave architecture, controlling direct and continuously the positions of the robot. For these tasks it is needed only for the system to have enough repeatability, once any absolute positioning errors can be visually compensated by the human operator. On the other side, in certain robotic applications, both programming techniques are either not practical or inefficient. The first, where the human operator teaches the trajectories to the robot, is not possible when there is high variance in the environment where the task is being performed. The second method, the teleoperation, has low absolute accuracy, due the deficiencies of human perception. This project follows modern tendencies of automation, which give increasing emphasis on robots guided by sensors and off-line programming, partially or completely automating many of the tasks to be executed. Sensors such as cameras eloed to a system of computational vision detect differences between the actual and desired position of the manipulator. This information is sent to controllers to correct the pre-programated trajectory. The manipulator movement commands are programmed off-line by a CAD system, without need even to turn on the robot, allowing for greatest speed on its validation, as well as problem solving. This work presents methodologies and techniques which allow the precise positioning of the manipulator using cameras in its end-effector. Once it is positioned in relation with the world frame, it is possible to move the manipulator with safety and precision its work area, as is needed for automation of complex tasks. This work is focused on computational vision techniques applied for manipulator calibration. It is based on a real case of a subsea manipulator of six degrees of freedom, used for underwater interventions in oil exploring platforms. The subjects treated in this work include camera calibration, pattern recognition, position tracking, stereopsis, direct and inverse manipulator kinematics and the union of all techniques for manipulator positioning in the work area.