Grasping force estimation detecting slip by tactile sensor adopting machine learning techniques

Adequate grasping force estimation and slip detection is a vital problem in wider applications of robots and manipulators in industries as well as in our everyday life. In this paper, a new methodology for slip detection during grasping by robot grippers/end-effectors using tactile sensor has been presented. During the object slippage, the tactile sensor in touch with the object surface travels along the peaks and valleys of surface texture of the object which creates vibratory motions in the tactile. A newly developed mathematical model is used to compute the scattered energy of vibrations, which contains parameters of surface texture geometry as well as trial grasping force, and other relevant parameters. Using the scattered energy of vibrations predicted by soft computing techniques, an attempt to instantly estimate the adequate grasping force has been reasonably successful. Surface texture data, for experimental estimation of grasping force, were collected from a huge number of machined specimens and were used to build four different machine learning estimation techniques. Experimental results using Linear Regression (LR), Simple Linear Regression (SLR), Pace Regression (PR) and Support Vector Machine (SVM) demonstrate a relatively better technique for industrial applications.