Touch recognition on FBG-based e-skins through deep and transfer learning

Artificial touch technologies can be used in robotics and healthcare to enable safe human-robot interaction and haptic feedback. For these goals tactile sensors can be embedded in soft e-skins, obtaining devices with wide and flexible sensitive areas. In this thesis Artificial Neural Networks (ANNs) have been developed to recognize and characterize touch in Fiber Bragg Grating (FBG) based e-skins, with a particular interest towards convolutive architectures. Transfer learning has also been adopted to reduce the amount of data required for model training and speed up their development.