A common approach to control multi-fingered grippers during stable grasping is the stiffness control scheme. "Internal" stiffness parameters and suitable references for internal grasping forces-often determined heuristically-ensure a stable grasp. In this paper, optimal internal forces are obtained in real-time by linearly constrained gradient flows on the smooth manifold of positive definite matrices. This optimization approach is a generic one for any number of fingers in contact with the object. Experiments with the 3-fingered Darmstadt hand show the simplicity and efficiency of the approach.

Whereas the emphasis of precision grasping only using the fingertips is placed on dexterity and sensitivity, power grasps guarantee high security and stability. But even if the task requires a power grasp, in the initial configuration the complete enveloping by the fingers is often impossible due to the position of the object in the workspace and only a fingertip grasp can be applied. Then a transition from precision to power grasping has to be performed. Based on an experimental study this paper discusses several possibilities to realize this transition by using a three-fingered dextrous hand mounted on a 6DOF robot. In particular it is distinguished between strategies requiring a coordination of robot and hand and strategies performing the transition only by the fingers like humans do.

This paper discusses the peg-in-hole task performed by a system integrating a dextrous hand, a 6-DOF robot and a camera. It is shown how the starting conditions of the final insertion only performed by the fingers are affected by the previous stages. Based on the experimental findings, some criteria for the overall system design and the required manipulation capabilities of the gripper are derived. With respect to this performance evaluation a modular concept for solving insertion tasks with a multifingered gripper is presented. The object manipulation is realized by primitives which are executed in dependence on various assembly states. The implementation of this approach is described and beside a theoretical characterization, experimental results are given.

This paper discusses couplings between the joints observed during position control using a tendon driven multifingered robot gripper. Depending on the mechanical construction these effects are non-linear and time variant as a consequence of friction and cable tension. Because the quality of the joint control is an important prerequisite for stable grasping, its design has to consider these coupling effects, generally using linear models. Therefore, the paper proposes a selftuning system which performs an iterative identification of a decoupling controller. The controller is stored in a CMAC-based neural network, so it can realize a non-linear decoupling. The authors present experimental results obtained by this method in comparison with a conventional approach.>