Learning stiffness characteristics of the human hand using a neuro-fuzzy system

Motor control of the human hand or of robotic manipulators is one of many i nteresting issues concerning motor learning. One possible way to achieve th is is by using stiffness as a control parameter. We built a system capable of learning and then successfully approximated the postural stiffness funct ion of the human hand over a horizontal workspace. The adaptive model built is based on a hybrid neuro-fuzzy system, called ANFIS. This system incorpo rates a fuzzy inference system based on the Takagi-Sugeno-Kang method, toge ther with a neural-network with learning capacities. The learning achieved has been used for fine-tuning the membership functions of the input. Simula tions carried out demonstrate the effectiveness of the hybrid system for po sture stiffness approximation of the human arm.