BeadSight: An Inexpensive Tactile Sensor Using Hydro-Gel Beads

TL;DR

BeadSight tackles the durability and cost barriers of visuo-tactile sensing by using a replaceable bead-based medium (hydro-gel PAM beads) encased in vinyl, observed by a rear camera. A U-Net learns to reconstruct contact pressure maps from multi-frame bead deformations, achieving an average MAE of $0.79\ \mathrm{kPa}$ and millimeter-scale localization accuracy. With a total device cost under $60 and cents-per-bead replacements, BeadSight offers a practical, robust tactile sensing solution for repetitive, high-contact robotics tasks, validated both in controlled experiments and a real-world apple grasp scenario. The work highlights potential integration with deep control policies to further improve robustness across domains.

Abstract

In robotic manipulation, tactile sensors are indispensable, especially when dealing with soft objects, objects of varying dimensions, or those out of the robot's direct line of sight. Traditional tactile sensors often grapple with challenges related to cost and durability. To address these issues, our study introduces a novel approach to visuo-tactile sensing with an emphasis on economy and replacablity. Our proposed sensor, BeadSight, uses hydro-gel beads encased in a vinyl bag as an economical, easily replaceable sensing medium. When the sensor makes contact with a surface, the deformation of the hydrogel beads is observed using a rear camera. This observation is then passed through a U-net Neural Network to predict the forces acting on the surface of the bead bag, in the form of a pressure map. Our results show that the sensor can accurately predict these pressure maps, detecting the location and magnitude of forces applied to the surface. These abilities make BeadSight an effective, inexpensive, and easily replaceable tactile sensor, ideal for many robotics applications.

Figures (8)

• Figure 1: BeadSight Overview. The sensor is composed of a camera mounted inside of a 3D printed housing, observing the deformations of hydro-gel beads encased in a vinyl pouch. A U-Net was trained to recreate contact pressure maps from the camera observations. During deployment, the bead sight can be mounted to a robot, and the recreated pressure maps can be used for applications, such as control.
• Figure 2: BeadSight hardware. The camera is mounted at the rear of a 3D-printed housing. At the front, a 40mm x 40mm acrylic window supports the hydro-gel bead pouch.
• Figure 3: Bead Bag fabrication process (a) 40 mm x 40 mm outline for drawn for edges. (b) 3 out of the 4 edges are heat sealed. (c) 100 PAM beads are inserted into the bag. (d) 5 ml of water is injected and (e) heat sealed. (f) Beads fully absorb water after 20 minutes.
• Figure 4: Architecture of the UNet model. The 15 prior frames are used to calculate the current pressure map.
• Figure 5: Visualization of experimental set-up. A 3D printer is used to create cylindrical presses at random locations and random depths. The force applied is recorded by a scale below the sensor. (Left) shows a graph of total force over time for a single press.