DIGIT: A Novel Design for a Low-Cost Compact High-Resolution Tactile Sensor with Application to In-Hand Manipulation

Mike Lambeta, Po-Wei Chou, Stephen Tian, Brian Yang, Benjamin Maloon, Victoria Rose Most, Dave Stroud, Raymond Santos, Ahmad Byagowi, Gregg Kammerer, Dinesh Jayaraman, Roberto Calandra

★ Must read May 2020

TL;DR The open-source, fingertip-sized GelSight-style sensor that made camera-based touch cheap and repeatable enough to put on multi-fingered hands. Shown driving learned tactile model-predictive control of glass marbles in hand.

What they do

Vision-based tactile sensors (GelSight and its descendants) image a soft elastomer from behind, so they give very high-resolution contact images. They were bulky, hand-built and fragile. DIGIT shrinks the design to a 20 g fingertip that mounts on an Allegro hand, simplifies manufacturing so many identical units can be built, and makes the gel last much longer. The design is fully open source.

How it works

A 60 fps camera on a custom USB PCB looks through an acrylic window at a pigmented silicone gel lit by three RGB LEDs. Contact deforms the gel and shows up as color and shading changes in the image.
The 20x27x18 mm body is three plastic pieces, 3D printed for prototypes or injection molded at scale. Camera and gel press-fit in, and the gel swaps out with one screw. The new gel holds up where earlier gels tore (Fig. 5 in the paper).
To show it is useful for control, they learn a tactile dynamics model (an autoencoder with a Gaussian-blob bottleneck) and run model-predictive control on the raw tactile images to roll glass marbles to target positions between two fingers.

Why it matters

DIGIT (and its commercial version from GelSight Inc.) became a default research tactile sensor: many later tactile-learning papers, simulators (TACTO) and datasets build on it. It is the reference point for the cost, size and resolution trade-offs that newer sensors argue against.

Limitations

Camera-based sensing makes the fingertip thick compared to skins. Normal and shear forces are not measured directly; they must be inferred from images.

Figures

Figure 1 : DIGITs mounted on an Allegro multi-finger hand. To validate our sensor design, we learn to manipulate glas... — Figure 1 : DIGITs mounted on an Allegro multi-finger hand. To validate our sensor design, we learn to manipulate glass marbles between two fingers.

Figure 2 : Exploded view of a single DIGIT sensor. A) elastomer, B) acrylic window, C) snap-fit holder, D) lighting P... — Figure 2 : Exploded view of a single DIGIT sensor. A) elastomer, B) acrylic window, C) snap-fit holder, D) lighting PCB, E) plastic housing, F) camera PCB, G) back housing.

Figure 3 : Object under test and corresponding raw measurements taken using DIGIT. The measurements taken from DIGIT ... — Figure 3 : Object under test and corresponding raw measurements taken using DIGIT. The measurements taken from DIGIT clearly capture sub-millimeters structures.

Figure 6 : System diagram of the self-supervised marble detector (top) and model predictive control using the learned... — Figure 6 : System diagram of the self-supervised marble detector (top) and model predictive control using the learned dynamics for marble manipulation (bottom). We first used the encoder part of the autoencoder network to detect the position of the marble from the tactile observations. We then trained a forward dynamics model predicting the position of the marble at the next time step, which we subsequently used to perform model predictive control. At each time step, an optimizer is used to find the best sequences of actions \mathbf{a}^{*}_{t:t+T-1} that moves the marble from the current position to the specified target position \mathbf{s}_{d} , and the first action \mathbf{a}^{*}_{t} is applied to the robot.

Abstract

Despite decades of research, general purpose in-hand manipulation remains one of the unsolved challenges of robotics. One of the contributing factors that limit current robotic manipulation systems is the difficulty of precisely sensing contact forces -- sensing and reasoning about contact forces are crucial to accurately control interactions with the environment. As a step towards enabling better robotic manipulation, we introduce DIGIT, an inexpensive, compact, and high-resolution tactile sensor geared towards in-hand manipulation. DIGIT improves upon past vision-based tactile sensors by miniaturizing the form factor to be mountable on multi-fingered hands, and by providing several design improvements that result in an easier, more repeatable manufacturing process, and enhanced reliability. We demonstrate the capabilities of the DIGIT sensor by training deep neural network model-based controllers to manipulate glass marbles in-hand with a multi-finger robotic hand. To provide the robotic community access to reliable and low-cost tactile sensors, we open-source the DIGIT design at https://digit.ml/.

Summaries are written automatically from each paper's text. Figures and videos come from the paper and its project page; all credit goes to the authors.

Matt Strong