One of the most important drawbacks of real robots is their lack of flexibility to changes in the environment or of the task. This paper presents a method enabling a visually guided robot to coordinate its movements with the vision processes. This method has been tested with a manipulator tracking maneuvering targets. The learning phase is performed in about one hour of repeated movements. After that, the robot can perform smooth and fast reaching movements and easily drop small objects into the wagon of a model train moving on an arbitrary trajectory.