A Neurofuzzy Knowledge Based Architecture for Robotic Hand Manipulation Forces Learning

Optimal distribution of forces for manipulation by a robot hand, is a hard computational issue, specifically once a whole hand grasp is needed. It becomes a complicated issue, once a robotic hand is equipped with human like deformable sensory touching materials. For computing optimal set of manipulation forces, grip transform and inverse hand Jacobian play major roles for such purposes. This manuscript is discussing a Neurofuzzy learning technique for learning optimal force distribution by a dextrous hand. For learning purposes, optimal set of forces patterns were gathered in advanced using optimization formulation technique. After that, to let a Neurofuzzy system to learn the nonlinear kinematics-dynamics relations needed for force distribution. This is done by considering the computational requirements for the inverse hand Jacobian, in addition to the interaction between hand fingers and the object. Training patterns clustering, and generation of the fuzzy initial memberships, and updated shape of memberships, are considered as vital information to build upon for more reasoning of fuzzy interrelation. The technique is novel in a sense, that the adopted Neurofuzzy architecture was transparent in terms of revealing the learned hand optimal forces if then rules.