A Machine-Learning-Based Approach to Solve Both Contact Location and Force in Soft Material Tactile Sensors.

This study addresses a design and calibration methodology based on numerical finite element method (FEM) modeling for the development of a soft tactile sensor able to simultaneously solve the magnitude and the application location of a normal load exerted onto its surface. The sensor entails the integration of a Bragg grating fiber optic sensor in a Dragon Skin 10 polymer brick (110 mm length, 24 mm width). The soft polymer mediates the transmission of the applied load to the buried fiber Bragg gratings (FBGs), and we also investigated the effect of sensor thickness on receptive field and sensitivity, both with the developed model and experimentally. Force-controlled indentations of the sensor (up to 2.5 N) were carried out through a cylindrical probe applied along the direction of the optical fiber (over an ∼90 mm span in length). A finite element model of the sensor was built and experimentally validated for 1 and 6 mm thicknesses of the soft polymeric encapsulation material, considering that the latter thickness resulted from numerical simulations as leading to optimal cross talk and sensitivity, given the chosen soft material. The FEM model was also used to train a neural network so as to obtain the inverse sensor function. Using four FBG transducers embedded in the 6-mm-thick soft polymer, the proposed machine learning approach managed to accurately detect both load magnitude (R = 0.97) and location (R = 0.99) over the whole experimental range. The proposed system could be used for developing tactile sensors that can be effectively used for a broad range of applications.