<p>Adaptive linkage-based underactuated hands are becoming popular because of their self-adaptability during power grasping without the need for complex sensing and control. Accordingly, they are widely used in unstructured spaces where it is difficult to use sensors to accurately obtain environmental and object information. However, owing to their nonlinear kinematics/dynamics and low repeatability caused by this same passive mechanism, it is challenging to actively and precisely control the robot state so that it can perform dexterous in-hand manipulation in the same manner. In this work, we make it possible for linkage-based underactuated hands with mechanical stoppers that were originally made for self-adaptive grasping to execute external sensorless in-hand object position manipulation in the hands’ full workspace. We first explain hand anatomy and how passive mechanisms cause nonlinear kinematics/dynamics and low repeatability in controlling the hand. Then, a suitable “virtual frame” in-hand object position manipulation technique is introduced and modified so that it can be used for high-level control. For the low-level control needed to compute the control signal of each actuator to perform the task, we estimate the fingertip force and position, which have low repeatability, from the fingers’ original internal force/position sensors via a novel method of clustering on a representative dataset and analytic modeling. In the real-world experiments, the novel framework enables a linkage-based underactuated hand to perform in-hand position control to its full ability with a wide range of objects.</p>