This paper focuses on the design of robust vision based control of a 2-DOF planar manipulator. The controller is expressed in the image space, where the system can be modeled as a pure integrator with a nonlinear matrix gain. It involves a simple proportional controller with a feedforward velocity input. Robustness is provided by the guarantee of a bounded tracking error, thanks to the online computation on the desired velocity along the trajectory to be tracked. The approach is both formally proven and experimentally validated on a lab apparatus.