GelBelt: A Vision-based Tactile Sensor for Continuous Sensing of Large Surfaces

TL;DR

GelBelt introduces a novel vision-based tactile sensor that uses an elastomeric belt sandwiched between two wheels to continuously scan large surface areas. By decoupling the elastomer from a rigid support and employing marker-assisted frame alignment, it enables high-resolution surface normal estimation and 3D surface reconstruction at speeds up to $45$ mm/s, with a reported average normal alignment of $ \{  \hat{\mathbf{n}}  \cdot \mathbf{n}_{ref} }  \approx 0.97 $. The workflow combines single-frame neural normal estimation from RGBXY inputs, optical-flow-based stitching, and Poisson integration to yield global meshes, while markers provide displacement, force, and angle cues for potential closed-loop control. Across single-frame and large-surface experiments, GelBelt demonstrates accurate surface reconstruction on small defects and larger areas, with a practical footprint and potential for handheld or autonomous deployment. This work has significant implications for automated quality control and maintenance in aerospace and other industries requiring rapid, large-area surface inspection.

Abstract

Scanning large-scale surfaces is widely demanded in surface reconstruction applications and detecting defects in industries' quality control and maintenance stages. Traditional vision-based tactile sensors have shown promising performance in high-resolution shape reconstruction while suffering limitations such as small sensing areas or susceptibility to damage when slid across surfaces, making them unsuitable for continuous sensing on large surfaces. To address these shortcomings, we introduce a novel vision-based tactile sensor designed for continuous surface sensing applications. Our design uses an elastomeric belt and two wheels to continuously scan the target surface. The proposed sensor showed promising results in both shape reconstruction and surface fusion, indicating its applicability. The dot product of the estimated and reference surface normal map is reported over the sensing area and for different scanning speeds. Results indicate that the proposed sensor can rapidly scan large-scale surfaces with high accuracy at speeds up to 45 mm/s.

Figures (7)

• Figure 1: Our GelBelt sensor is mounted on a UR5e robot for continuous surface reconstruction of an aircraft part to detect defects. Gelbelt can potentially be motorized to scan the surface on its own.
• Figure 2: GelBelt's mechanical design. (A) Highlighted optical components. (B) CAD exploded view showing all the components in the model.
• Figure 3: GelBelt's optical configuration. (A) We enhanced the optical system by changing the light locations in simulation to improve the light intensity and contrast over the entire sensing area, which highly matched the real sensor image. (B) A simplified representation of the light placement. The thin air gap between the belt and acrylic results in total internal reflection.
• Figure 4: GelBelt's single-frame sensing capability on random objects. From left to right: Chinese Yuan bill, screw, coin, scotch tape, and a key.
• Figure 5: GelBelt reconstruction accuracy. (A) We indent a hex pyramid shape in multiple locations and obtain the 2D normal estimation accuracy plot. (B) We roll over the same object to obtain the accuracy for different offsets from the image centerline. (C) The effect of rolling speed on the sensor accuracy. (D) 3D mesh reconstruction comparison to GelSight Max, a commercialized VBTS with nanometer scale z-axis accuracy.