DIC-Gel: a digital image correlation-enhanced neural network-based tactile sensor for multi-parameter contact estimation

Vollständiger Abstract

Worum geht es in dieser Arbeit?

Abstract Vision-based tactile sensors (VBTS), widely adopted in robotic end-effectors for tactile cues to support environmental perception and interaction, have attracted significant attention due to their structural simplicity, high spatial resolution, and robustness. However, state-of-the-art VBTS over-rely on complex optical configurations and cumbersome traditional markers, limiting their miniaturization, integration, and practical deployment. To address these limitations, this work proposes a novel VBTS design framework integrating digital image correlation (DIC) with deep learning, and develops a tactile sensor prototype termed DIC-Gel. Unlike conventional designs, DIC-Gel uses a speckle-based surface pattern as intrinsic markers, enabling real-time capture and quantification of subtle contact interface deformations with high resolution. With this design, DIC-Gel achieves synergistic optimization of multi-parameter tactile estimation: rapid response without complex optical arrangements, reducing system complexity and cost. Experimental results show a normal force measurement error of 0.24 N, lateral coordinate errors of 0.69 mm ( X/Y axes), and a depth displacement error of 0.03 mm, indicating its application potential in robotic manipulation, intelligent grasping, and human-robot interaction.