Source Data.zip

Soft mechanical sensors with extreme mechanical robustness, high performance, and manufacturing reproducibility are crucial for robotics perception, but simultaneously satisfying these criteria is rarely achieved. Here, we suggest a magnetic crack-based piezoinductive sensor (MC-PIS) which exploits the strain modulation of magnetic flux in cracked ferrite films. The MC-PIS is insensitive to fatigue-induced crack propagation and environmental changes, showing same performance even when scratched in half or run over by a car. It can detect bidirectional bending with a precision of 0.01° from -200° to 327°, allowing for real-time reconstruction of dynamic shape changes of a flexible ribbon. We demonstrate an artificial finger recognizing surface topology and musical notes via vibrations, a crawling robot responding appropriately to external stimuli, a tree-planting gripper performing consecutive tasks from digging soil, removing stones, to placing trees. The MC-PIS opens a new paradigm to develop ultrasensitive yet extreme robust sensors in real-world robotics applications.