FBG Tactile Sensing System Based on SVP-Transformer for Material Classification

Tactile is an essential sensory modality for the interaction between the human hand and the surrounding environment, playing a crucial role in tasks such as object grasping, tactile perception, and object classification. In this study, we propose a material classification scheme based on fiber Bragg grating (FBG) tactile sensing technology and deep learning algorithms. Considering the sensing characteristics of FBGs and the principles of human tactile dynamic and static perception, the biomimetic finger is equipped with FBGs, which is based on the simulation results of ANSYS finite element. By simulating the sliding motion of biological fingers, tactile data is collected from different material surfaces. Wavelength-swept optical coherence tomography technology is used to obtain the interference signals of the FBGs, which are then processed to obtain the sensing signals of static stress and dynamic vibration. Based on the obtained dynamic and static multivariate time series tactile datasets, we construct a Transformer-based multivariate time series classification model named Shape-level Variable-Position Transformer (SVP-T) model, for the classification of eight common object surface materials. The average classification accuracy reaches more than 86%, and the single recognition time is only about 0.94 s, which proves the effectiveness of the combination of sensing system and deep learning algorithm. The method proposed in this article has broad application prospects in the fields of healthcare, robotics, and industrial manufacturing.