Deep Learning-Assisted Intelligent Liquid Crystal Elastomer Grippers Based on Autonomous Triboelectric Sensing

Soft grippers have shown promising applications in robotics due to their high flexibility, damage-free contact, and environmental adaptability. However, their sensing often relies on external sensors and thus suffers from susceptibility to environmental interference. Here, we report a liquid crystal elastomer (LCE) gripper integrated with dual-mode triboelectric nanogenerators (TENGs) for self-powered target identification. By synergizing fluorinated ethylene propylene (FEP) and polydimethylsiloxane (PDMS) TENG sensors, the system generates voltage signals (V₁, V₂) encoding intrinsic material properties and kinematic parameters during object interactions. The hybrid convolutional neural network-long short-term memory (CNN-LSTM) architecture extracts discriminative spatiotemporal features from raw triboelectric/electrostatic signatures, achieving 94.4% classification accuracy across 5 material categories through cross-validation. This fusion of contact electrification physics and deep learning overcomes traditional limitations in environmental interference susceptibility, establishing a paradigm for perceptually intelligent soft robotics in industrial automation and human-machine interaction scenarios.