TouchRoller: A Rolling Optical Tactile Sensor for Rapid Assessment of Textures for Large Surface Areas

Guanqun Cao, Jiaqi Jiang, Chen Lu, Daniel Fernandes Gomes, Shan Luo

TL;DR

This paper introduces the "TouchRoller," a novel rolling optical tactile sensor designed for rapid texture assessment of large surfaces. By enabling continuous rolling contact, it significantly outperforms traditional flat tactile sensors in speed and efficiency, while maintaining high-resolution tactile data collection and accurate contact localization.

What Problem Does This Paper Address?

Robots need tactile sensing to understand their environment, but current tactile sensors, especially optical ones, often have small sensing areas. This necessitates numerous repetitive actions like pressing, lifting, and shifting to cover large surfaces, making the process slow and inefficient. Furthermore, dragging these sensors can damage them or the object. The paper addresses the challenge of efficiently and effectively gathering detailed tactile information from extensive surface areas, a crucial capability for tasks like robotic inspection and manipulation.

Methods

The core innovation is the "TouchRoller," a sensor with a cylindrical design that allows it to roll. It features a transparent elastomer covering a reflective membrane on a transparent cylinder. An internal camera captures the deformation of the elastomer as it contacts a surface. This deformation, illuminated by internal LEDs and reflected by the membrane, provides a high-resolution tactile image. The cylindrical shape allows for continuous rolling, maintaining stable contact and enabling efficient data acquisition across large areas. The paper details the sensor's fabrication process, including the elastomer mixture and reflective membrane application, and outlines a surface projection method to map 3D contact points to 2D tactile images. Calibration procedures for intrinsic camera parameters and extrinsic transformations are also described.

Datasets

The paper does not explicitly mention training datasets. For evaluation, they used a piece of fabric with an 8 cm x 11 cm textured surface. They also used a 3D-printed solid with hemispheres for contact localization experiments and a chessboard pattern for camera intrinsic calibration.

Key Insights and Ideas

  • Rolling for Efficiency: The fundamental insight is that a rolling motion, rather than repeated pressing and lifting, dramatically speeds up tactile data acquisition for large surfaces.
  • Continuous Contact: Maintaining continuous contact during rolling leads to more stable data and avoids the issues of slipping and dragging associated with flat sensors.
  • High-Resolution Texture Mapping: The optical nature of the sensor, combined with the deformation of the elastomer, allows for detailed reconstruction of surface textures.
  • Integrated Localization: The sensor design inherently supports the localization of contact points, which is vital for robotic manipulation and exploration.

Key Results

  • Speed: The TouchRoller covered an 8 cm x 11 cm textured surface in just 10 seconds, a significant improvement over a flat optical tactile sensor which took 196 seconds for a similar task.
  • Texture Mapping Quality: Reconstructed texture maps achieved an average Structural Similarity Index (SSIM) of 0.31 when compared to visual texture, indicating good quality.
  • Contact Localization Accuracy: The sensor demonstrated low localization errors, with an average error of 2.63 mm in the center regions and 7.66 mm on average across the entire sensing area.
  • Robustness to Speed: The sensor performed well at modest rolling speeds (around 1.1 cm/s), producing clear tactile images.

Limitations and Future Work

The authors acknowledge several limitations:

  • Size: The current sensor is relatively large, making it unsuitable for confined spaces or measuring curved surfaces.
  • Dexterity: Its integration as an end-effector limits its application in highly dexterous manipulation tasks.
  • Durability: The soft elastomer is delicate and susceptible to wear from frequent contact or high temperatures.
  • Maintenance: Regular re-calibration and maintenance are required, which can be costly and time-consuming.

Future work will focus on optimizing the fabrication process, developing more durable materials, reducing the sensor's size for easier integration with robot grippers, and exploring applications in robotic manipulation.

Summary updated

Abstract

Tactile sensing is important for robots to perceive the world as it captures the physical surface properties of the object with which it is in contact and is robust to illumination and colour variances. However, due to the limited sensing area and the resistance of their fixed surface when they are applied with relative motions to the object, current tactile sensors have to tap the tactile sensor on the target object a great number of times when assessing a large surface, i.e., pressing, lifting up, and shifting to another region. This process is ineffective and time-consuming. It is also undesirable to drag such sensors as this often damages the sensitive membrane of the sensor or the object. To address these problems, we propose a roller-based optical tactile sensor named TouchRoller, which can roll around its centre axis. It maintains being in contact with the assessed surface throughout the entire motion, allowing for efficient and continuous measurement. Extensive experiments showed that the TouchRoller sensor can cover a textured surface of 8 cm × 11 cm in a short time of 10 s, much more effectively than a flat optical tactile sensor (in 196 s). The reconstructed map of the texture from the collected tactile images has a high Structural Similarity Index (SSIM) of 0.31 on average when compared with the visual texture. In addition, the contacts on the sensor can be localised with a low localisation error, 2.63 mm in the centre regions and 7.66 mm on average. The proposed sensor will enable the fast assessment of large surfaces with high-resolution tactile sensing and the effective collection of tactile images.
BibTeX
@inproceedings{cao2023touchroller,
  author = {Guanqun Cao and Jiaqi Jiang and Chen Lu and Daniel Fernandes Gomes and Shan Luo},
  title = {{TouchRoller: A Rolling Optical Tactile Sensor for Rapid Assessment of Textures for Large Surface Areas}},
  booktitle = {Sensors},
  year = {2023},
  url = {https://openreview.net/forum?id=Fli2Q00hmr},
}
Figure