The work deals with the development of the control of an articulated mechanical hand with the aim of fine manipulation of objects. This hand possess four fingers with four degrees of freedom. The sixteen joints are actuated using cable transmission and sixteen DC servo actuators. Position and force controls as well as their performances are presented and developed through examples. For the force control, an original method based on learning is developed. This method, using neural networks, allows us estimate the contact force by modeling the behaviour of the elastic transmission made with cables.

This paper deals with optimization of grasping with multi-fingered robot hands, under general constraints such as finger deformability, sliding conditions and object positioning tolerances. A general formalism describing hyperstatic grasping is presented. An optimization criterion based on the minimization of securing forces and torques is introduced. Results concerning numerical simulation of grasping with a three-fingered gripper are presented.>

The authors describe an efficient and fast algorithm for finding the minimum distance between two convex polyhedrons in three-dimensional space. To obtain the minimal distance, the proposed computational scheme is based on a direct approach to minimize the distance function which produces a succession of optimal search directions along the boundary of the objects. This algorithm combines the gradient projection method (Rosen, 1960) and an additional optimal search direction when the gradient projection method leads to a zigzagging phenomenon. In this case, the additional optimal search direction accelerates significantly the convergence of the process. Extensive numerical experiments with convex polyhedra showed the performance of the algorithm when compared with previous approaches.>

Dynamic optimization of the motion of a robot arm is carried out using the Pontryagin maximum principle. A mixed performance index involving travel time and bounded actuators is minimized to obtain a near-minimum traveling time and continuous output. One can avoid, in that case, actuating effort jumps and arm jerks. A full dynamic model formulated with Hamiltonian variables is used. This model, together with the optimality conditions, leads to a two-point boundary value problem which is solved by combining a gradient algorithm and a shooting method.>