Transfer learning for vision-based tactile sensing

Abstract

Due to the complexity of modeling the elastic properties of materials, the\nuse of machine learning algorithms is continuously increasing for tactile\nsensing applications. Recent advances in deep neural networks applied to\ncomputer vision make vision-based tactile sensors very appealing for their\nhigh-resolution and low cost. A soft optical tactile sensor that is scalable to\nlarge surfaces with arbitrary shape is discussed in this paper. A supervised\nlearning algorithm trains a model that is able to reconstruct the normal force\ndistribution on the sensor's surface, purely from the images recorded by an\ninternal camera. In order to reduce the training times and the need for large\ndatasets, a calibration procedure is proposed to transfer the acquired\nknowledge across multiple sensors while maintaining satisfactory performance.\n