To ensure optimal performance of parallel robots, a rigorous design process has to be implemented. However, this may not be enough due to the presence of complex phenomena such as vibration, clearance, deformation, hard to model and thus to control, but considerably impacting the robot performance. An efficient approach to improve performance via bypassing the modeling issues is the use of exteroceptive sensors to estimate the end-effector pose. Any external observation, however, impacts the robot performance. It is thus necessary to optimize the robot design with respect to (usual) mechanical performance criteria, but also with respect to performance indices coming from the definition of the sensor-based controller. Thus it is necessary to achieve control-based design. In this work, a five-bar mechanism is optimized using a classical design methodology. The robot is then compared with other designs which are the result of a new methodology, taking into account the nature of the desired control scheme through the incorporation of control-based performance indices into the optimization process. Though the latter designs may have a bigger footprint, they will prove to exhibit better accuracy performance when controlled using exteroceptive sensors.