A Bioinspired, Multimodal Soft Tactile Skin with Task-Adaptive Perception for Intelligent Robotic Manipulation

Achieving human-like tactile perception is essential for robotic systems to perform fine manipulation and adapt to dynamic environments. However, most tactile sensors have deviated from the physiological encoding based on human mechanoreceptors. Here, we present a layered trimodal tactile sensing system that mimics Meissner, Merkel, and Ruffini receptors. These sensors can accurately distinguish stimuli (static normal pressure, dynamic shear vibration, static horizontal strain) coming from different directions and in different shapes. Using a receptor-level analysis framework based on random forest feature importance, we systematically analyze the involvement and relative contribution of biomimetic sensors across four tactile tasks (Braille reading, texture identification, softness classification, slip detection). Notably, the cooperation of two or three sensors significantly enhances recognition accuracy for the tasks. When the sensors are integrated on a robotic gripper, they enables real-time slip/drop detection with 97.62% accuracy and supports closed-loop grip-force adaptation, thereby demonstrating stabilized grip of the object under increasing load. These results help identify appropriate sensor combinations for specific tactile tasks and demonstrate the potential of the proposed system for applications in humanoid robotics, prosthetics, haptic devices, and augmented and virtual reality systems.