To achieve reliable grasping of unknown objects by robots and address the problems of existing tactile sensors

such as narrow force range and low durability

a slip detection method fusing vision

optical flow

and tactile sensing is proposed.A flexible ionic tactile sensor array characterized by high load capacity (0~100N)

high sensitivity (156.6kPa

-1

)

and long service life (>150000 cycles) is developed.A multimodal dataset containing 6400 annotated sequences is constructed

in which tactile data is innovatively characterized as spatiotemporal graphs.The proposed triple modality fusion with graph attention network(TFG-Net) model is designed for multimodal feature ex

traction and hierarchical fusion.Experimental results show that the model achieves a slip detection accuracy of 91.76% on unknown objects

which outperforms mainstream sequence models.Closed-loop experiments verify a success rate of 87.0% in slip recovery.