The vision of robotic assistants for domestic, health care, and workplace applications will not come to fruition without the ability to manipulate typical objects in human environments. However, grasping is challenging in unstructured environments because object models typically required to control the robot are not known beforehand and must be acquired through sensors that are imprecise and incomplete. The majority of research in robotic grasping and manipulation has attempted to address this problem through elaborate multifingered hands combined with tactile sensing and sophisticated planning and control algorithms, often following an anthropomorphic approach. This proposal utilizes an alternative approach involving a focus on the mechanics of the hand itself to accomplish most of the needed ?control.? By appropriately incorporating features such as compliance and underactuation, the uncertainty inherent in unstructured grasping tasks can be more easily accommodated. The proposed work addresses the problem of precision grasping of small objects from the surrounding environment and then begins to address the broader problem of dexterity by examining two-fingered precision manipulation, while investigating the role of compliance, underactuation, and configuration on performance in the presence of uncertainty. This work will be disseminated through publications in scholarly journals and conferences and will contribute to the fundamental understanding of the mechanical interaction of robot hands with small objects and the surrounding environment. These results are expected to lead to the development of low-dimensional hands for precision robotic grasping and manipulation with applications including assistive robots and prosthetics.