This paper deals with the problem of unavailability of velocity measurements for the control of parallel manipulators for pick-and-place applications. As most controllers require the measurement of the joint positions as well as joint velocities, the latter need to be estimated. Three observers have been implemented while keeping the same nonlinear controller: a Lead-lag filter based observer, an Alpha-beta-gamma observer and a High-gain observer. The resulting performances obtained in Real-time experiments by each observer have been detailed and compared.