In order to improve the sensitivity and accuracy of the tactile perception of the fingertip of the machine, based on the analysis of the mechanism of the fingertip tactile perception machine, a feature-separated double-layer “cross”-type FBG tactile sensing unit was designed, and finite-element simulation analysis was carried out, and calibration experiments and grip sensing experiments were carried out for the sensing unit. Based on the composite perception of contact temperature and grip force, a coupling analysis was carried out, and a decoupling method based on whale optimization algorithm for optimising back propagation neural network ( WOA-BPNN) is proposed. The experimental results show that the contact temperature sensitivity of the FBG sensing unit is 11. 255 pm/ ℃ , and the grip force sensitivity is 17. 342 nm/ MPa; the average absolute error of the contact temperature of the WOA-BP decoupling model is reduced by 72. 53% , and the average absolute error of the grip force is reduced by 68. 55% .