A neural network architecture to learn hand posture definition

The goal of our work is to propose some solutions to hand posture definition and learning process during object manipulation and grasping. The main idea is to try to transfer relevant central nervous system strategies and related parameters obtained by experimentation to the frame of artificial neural networks in order to perform the learning of the underlying brain functions in the field of prehension. As a first step toward this goal and in order to evaluate and simulate hand manipulation capabilities, it is necessary to define means to evaluate all the relevant parameters associated to hand geometry and kinematics. In this frame, we propose to study two particular aspects that are hand geometry through the development of a hand "morphologic generator" able to build an arbitrary hand model and a finger posture determination algorithm based on a modular neural network architecture (called Hand Posture Modular Neural Networks Architecture or HP-NNA).