This paper describes how a forward model could be applied in a manipulator robot to accomplish a reaching task. The forward model construction is optimized using validity indexes. The forward model has been implemented in the puma 560 robot manipulator in simulation and deals with the occlusion problem. So the robot can start with its end effector in any random position of the workspace, the forward model will find a position in which the end effector will be visible in the robot image plane using a Cartesian controller. Once the end effector is in a visible position an image based visual servoing algorithm is implemented for a reaching task. Results and simulations are shown to demonstrate the applicability of this proposed architecture for a visual servoing task. © 2013 Praise Worthy Prize S.r.l. - All rights reserved.