A hybrid model for the hand preconfiguration in rehabilitation grasping tasks

This paper addresses a hybrid method for adapting a three-fingered dextrous and compliant artificial hand, in order to describe preconfiguration before grasping some specified rigid object. For that, we use an anthropomorphic model based on shape memory alloy actuators with a neural control architecture. This model integrates neural and symbolic approaches to biomechanical hands. The hybrid characteristics of both methods and the modularity of control algorithms allow one to relate different kinds of kinematic models, learning and control processes. We have developed a variant of DIRECT model based on difference vectors, which does not require a prior knowledge of the geometry of the manipulator and makes it easier to control velocities in the joints near mathematical singularities