Neural Network Approach to Stiffness Based Touch Sense Storage and Reproduction

In this paper, a sliding mode neural network is utilized to learn environmental conditions during haptic touch of bilaterally controlled robot to an unknown environment. Learning of environmental conditions is based on obtaining the highly nonlinear data mapping between force and position dimensions by the neural network. The environment identifier network is then utilized to reproduce the environmental conditions in the absence of the environment. The exact feeling of touch is reproduced by means of environmental conditions. Real time experiments on haptic forceps robot that is controlled by a hybrid force-position controller are carried out to verify the viability of neural network approach to recording and reproduction of haptic touch sense which is based on evaluation of stiffness.