StRETcH: A Soft to Resistive Elastic Tactile Hand

Authors: Carolyn Matl, Josephine Koe, Ruzena Bajcsy · ICRA 2021

Licences

Open Access
Paper: CC BY 4.0

Everything below draws on the full paper under its CC BY 4.0 licence; its figure is credited to the authors.

TL;DR

An EcoFlex membrane stretched by a parallel-jaw gripper and imaged by a depth camera gives a soft tactile end-effector with adjustable stiffness for sensing and shaping deformable objects.

Why it matters

The design reuses a standard parallel-jaw gripper and an off-the-shelf depth camera, with a cast silicone sheet and 3D-printed finger mounts as the only custom parts. One end-effector can switch between a soft, conformable mode for measuring geometry and a stiffer mode for pressing and shaping objects, which the dough-shaping demonstration uses in sequence.

Problem

Manipulating solid deformable objects such as dough needs both contact geometry and the object's compliance, but conventional tactile sensors usually measure one or the other. Soft optical sensors such as GelSight image contact in high resolution but have a fixed membrane stiffness, so they sense only a narrow range of object hardness before deforming the object themselves.

Proposed solution

StRETcH suspends a flat 100 × 90 × 2 mm EcoFlex 50 silicone membrane between the fingers of a Robotiq 2-finger gripper on a UR5 arm, with an Intel RealSense SR305 depth camera imaging the membrane from behind. Opening the gripper stretches the membrane uniaxially and raises its stiffness; contact regions are segmented from the depth image by local curvature. The membrane's response is modelled with Hertz contact theory as an equivalent elastic half-space whose effective modulus E* is calibrated against stretch, which lets the hand estimate contact force from contact geometry and stretch state.

Key insights

Uniaxially stretching the membrane with the gripper is an easier way to control stiffness than the pneumatic pressurisation used in the authors' earlier SOFTcell fingertip.
Low stretch states give the highest geometric sensitivity, while higher stretch states allow larger contact forces, so the hand works as a multimodal sensor-actuator.
The stiffer the membrane, the deeper the indentation needed for sub-millimetre geometry accuracy; indentations under 2 mm performed poorly for all test cylinders.
Treating the membrane as an equivalent elastic half-space is not physically exact but fits the force data closely beyond about 3 mm indentation and is fast enough for real-time use.
Measuring the contact radius first, then raising stretch while pressing until the object deflects, gives a stiffness proxy that separates materials of different compliance.

Main results

Effective modulus is controllable from about 4 kPa to 9 kPa by stretching the membrane, with low variance over 5 trials.
Contact geometry estimates reach sub-millimetre accuracy under suitable indentation depth and stretch; wooden-object reconstructions were within 2 mm of real dimensions and paper object heights within 3 mm.
Leave-one-out load-force estimation error averaged 0.023 ± 0.011 N, under 1% of the operating range (Section V-C); the conclusion states the average error as less than ±0.025 N.
Stiffness proxies for balloons: air 0.387 ± 0.081, water 0.726 ± 0.085, chia pudding 1.112 ± 0.178 and chia seeds 1.431 ± 0.182 N/mm.
Play-Doh cubes were rolled into spheres and pressed into cookies with average circularity 0.888 (0.785 is a square, 1 a circle).

Limitations

Stiffness, and so the force exerted, is upper-bounded by the membrane thickness and the maximum stretch the gripper allows.
The Hertz half-space model is an approximation; a membrane is mechanically different and more complex than an elastic half-space.
Contact segmentation assumes objects are flat or convex and larger than a single surface element of the membrane.
Rolling dough into a sphere took about ten minutes on average and up to twenty; tangential traction during rolling is not modelled.
The balloon stiffness values are proxies from a one-degree-of-freedom approximation, and depth-sensor noise was about ±1 mm.

Field impact

The paper won the IEEE ICRA 2021 Best Paper Award in Robot Manipulation and has 5 citations. It describes the full build (membrane casting, mounts, sensor) but does not state a code, data or design-file release.

Abstract

Soft optical tactile sensors enable robots to manipulate deformable objects by capturing important features such as high-resolution contact geometry and estimations of object compliance. This work presents a variable stiffness soft tactile end-effector called StRETcH, a Soft to Resistive Elastic Tactile Hand, that is easily manufactured and integrated with a robotic arm. An elastic membrane is suspended between two robotic fingers, and a depth sensor capturing the deformations of the elastic membrane enables sub-millimeter accurate estimates of contact geometries. The parallel-jaw gripper varies the stiffness of the membrane by uniaxially stretching it, which controllably modulates StRETcH’s effective modulus from approximately 4kPa to 9kPa. This work uses StRETcH to reconstruct the contact geometry of rigid and deformable objects, estimate the stiffness of four balloons filled with different substances, and manipulate dough into a desired shape.

Open Source Robotics