In this work, we address robotic dexterous ungrasping, the capability of securely transferring a grasped object from the gripper to the robot’s environment, i.e. the inverse of grasping or picking, through dexterous manipulation. The situation of dexterous ungrasping can commonly be found in many occasions such as placing a Go stone or inserting a battery pack, where it is necessary to carefully perform contact-rich interactions for stable ungrasping. Advancing from our recent work that formulated a planning framework for dexterous ungrasping, this study addresses the robustness of the motion control for executing the planned ungrasping operations, through an analysis of possible failure cases and caging conditions. We also address how the guaranteed robustness facilitates the modeling and planning of dexterous ungrasping and guides the arrangement of a gripper system. A series of experiments demonstrate the effectiveness of the resulting hardware and software system in practical ungrasping tasks.