A telemanipulation system lets a human operator manipulate an environment on distance, using a master and a slave device, which are connected by a bilateral controller that controls the master and the slave device and provides the communication between these devices. The master device is manipulated by the operator and these actions are transferred to the slave device, using the bilateral controller. The interaction forces between the slave device and the remote environment are fed back to the operator, which provides awareness and allows a task to be executed efficiently and safely. Currently, one of the limitations of telemanipulation systems is that interaction forces are typically indirectly transferred to the human operator. Firstly, a commonly used concept to determine the feedback forces, is by coupling the master and slave position, using a proportional controller. Secondly, feedback forces are generally applied on the master device, which then transfers these forces to the operator through its end effector. Because of the dynamics of the master device, these forces are distorted as they are transferred to the operator. This reduction in force feedback quality decreases awareness of the operator and thereby likely reduces his or her task performance. To minimize the distortion of the feedback forces by the dynamics of the master device, often a parallel manipulator is used as master device. Compared to a serial manipulator, a parallel manipulator has the advantage that the moving masses are small, because the actuators are positioned at the base. However, the difficulty of parallel manipulators is the kinematic and dynamic modeling of the device, which is much more complex than the modeling of serial manipulators. The goal of this research is to improve force feedback quality. This is achieved through the design of a local admittance controller for a parallel master device, which will cancel out the dynamics of the parallel master device and allows them to be replaced by desired dynamics (desired mass, damping and stiffness). Using this approach, the feedback quality will not only be of higher quality, but will also enable a more complete and intuitive method to design the force feedback. However, generally an admittance controller requires the external force applied by the operator as input. Force sensors are expensive and fragile. As a more cost-effective solution, in this thesis the external force will be obtained by force estimation using a model based observer. The model based observer only requires the position measurements, and control wrench (combination of force and torque) at the end effector, as input. The admittance controller for the parallel master device, without the use of a force sensor, is configured by combining insights from different fields of robot analysis and control. The control algorithm will be build up by a computed torque controller, the desired dynamics to shape desired acceleration, an observer, an observer based force estimator and methods to derive a kinematic and a dynamic model of a parallel manipulator. The control method is tested on a three degrees of freedom planar parallel device: the Munin. The tests show that the algorithm is able to replace the dynamics of the master device with desired dynamics, such that the device admits a certain motion, with respect to the desired dynamics and the external force applied by the human operator. The limits of the range of desired stiffness, damping and mass, that can be projected on the manipulator’s end effector by the control algorithm, are explored. This range is large enough to conclude that the algorithm is suitable to be used on parallel master devices, to project desired dynamics on the end effector of the device, without the use of a force sensor. The control algorithm is designed to control the master device of a telemanipulation system. This enables a intuitive configuration of force feedback. The desired dynamics can be shaped, which can be beneficial to advanced control concepts, such as haptic shared feedback and model mediated control, on top of the feedback directly available form the measurements at the slave device. Furthermore, the control algorithm could also be beneficial for the safe operation for all kinds of devices that cooperate with a human, because of the force estimation, instead of force measurement on one point of the device, will react to all forces applied anywhere on the device.