Abstract: Conventional tactile sensing platforms face persistent challenges in interpreting complex spatiotemporal tactile signals. Herein, we report a next-generation tactile system combining a mechanically adaptive pressure sensor array with a deep learning-based multimodal neural network. The pressure sensor is designed and fabricated by constructing a silver nanoparticles (AgNPs)/reduced graphene oxide (rGO) dual-conductive network on a compressible porous melamine foam (MF) via dip-coating and in-situ reduction. It exhibits tunable sensitivity (0.025 to 0.277 kPa^-1) within a pressure range of 0-52.28 kPa, fast response (120 ms), and high durability with < 3 % signal fluctuations after 3000 cycles. Importantly, the platform features a dual intelligent decoding architecture: (1) a CNN-based spatial analyzer achieves 98.52 % classification accuracy for six object types, surpassing conventional thresholding methods; (2) a Bi-RNN temporal decoder enables 100 % authentication accuracy in haptic sequence-based password recognition. Unlike human haptic cognition which requires multisensory integration, our system leverages machine learning for efficient interpretation of tactile information. This work paves the way for advanced human-machine interfaces in assistive robotics and virtual reality, particularly enhancing tactile learning for visually impaired users.

Key words: Tactile sensing platform, Pressure sensor, Deep learning, Intelligent recognition