Nanocrack Encoded Electronic Skin for Vectorial Strain Sensing

Human skin perceives both the magnitude and direction of mechanical stimuli, an essential capability for dexterous interaction yet fully unrealized in humanoid robotics. Here we introduce a metallic nanocrack-based electronic skin (Crack-eSkin), a strain vector perceptron that encodes the complete strain vector through the algorithmic geometry of nanocracks in the 50 nm thick gold film. When mechanically stretched, orthogonal cracks open differentially depending on orientation, producing distinct signals that simultaneously encode both strain magnitude and direction as a two-dimensional resistive fingerprint. This mechanism transforms mechanical stretch into an ultra-sensitive (gauge factor: 3609) material-level response that preserves both parameters at their source. Feeding encoded signal sequences into the developed geometry-aware multitask temporal network (RMT-Net) enables the joint prediction of strain magnitude and loading angle, achieving R² values of 0.982 for strain and 0.985 for angle. Installed onto an artificial face model, the Crack-eSkin can quantitatively resolve sequential strain vectors of 2.60% at 87.4°, 1.84% at 7.7°, and 3.20% at 45.2° that human subjects could only describe qualitatively, paving the way for next-generation robotic systems with human-like or superior mechanosensation.