A neural estimator of object stiffness applied to force control of a robotic finger with opponent artificial muscles

We present a solution for real-time neural estimation of the stiffness characteristics of objects which are pressed with a predefined force threshold by an anthropomorphic robotic finger provided with opponent movement of their artificial muscles. The proposed architecture links three neural models in order to satisfy the requirements in our control system. This model based on adaptive learning allows the controller to grasp any object with different stiffness characteristics in a smooth way and with the desired final force.