A $5 Fabric Robot Grips Objects From a Human Hair to a Vase

A single human hair, roughly 90 micrometers across, is being held steady by something that looks like a party blower made of coated fabric. The gripper doing the holding weighs about 9 grams. Push air into one of its two chambers and it uncurls; let the air out and a flattened coil spring inside drags it back into a tight planar spiral. The same object, minutes later, wraps itself around a vase.

Robotic hands have spent decades chasing the human original bone for bone and tendon for tendon, and the results are usually heavy, pricey and a nightmare to control. The team reporting in Advanced Science decided to skip the anatomy and copy the jobs instead.

Copy the Job, Not the Hand

Their robot, which they call the BioflexBot, is two fabric air pouches bonded onto that spring. Inflate the pouch running along the top surface (it is split into two channels, which stops it ballooning upward) and the spring straightens, so the spiral extends and rotates as it unwinds. Inflate the pouches down the sides instead and the whole thing bows outward into a wide C, ready to swallow something big. Slip a rigid shell over it to suppress that sideways bulge, and the same inflation closes the inner cavity rather than opening it, which is what turns the machine into a pincer.

Three motions, then. Extension, expansion and contraction, all of them running off two pneumatic inputs, which is really the number that matters.

Because the range of things a gripper can pick up scales with how far it can open and close, that spiral does something a jointed hand cannot. It stretches to about seven times its contracted diameter and pulls back again, a ratio the authors put at roughly 3.5 times what a human hand manages. Set against published figures for other grippers with a comparable number of actuators, the span of object sizes it handles comes out more than 12.9-fold wider.

“Unlike most robotic hands that replicate the human form, at high hardware and control costs, our approach focuses solely on mimicking the functions, not the shape,” says Yang Yang at Nanjing University of Information Science and Technology, one of the study’s senior authors. His co-senior author Yingtian Li, now at the Chinese University of Hong Kong, Shenzhen, describes the aim as a design simple enough to build cheaply and still able to grasp across scales.

Beyond the Reach of One Hand

Some of what follows is straightforwardly beyond a hand. The robot grips the inner walls of cavities from 30 to 210 mm across, unscrews a bottle cap by turning 690 degrees in one continuous motion (a human hand manages about 180 before you have to let go and regrip), posts itself through a 35 mm slit to retrieve a pair of scissors sitting 135 mm inside a box, and holds a glass container and a pen at the same time before setting each one down in a different place. A geometric model predicted the largest object each bow shape could enclose, and the measured values tracked it. There is fine work too: the pinching mode guided a 1.1 mm acupuncture needle into a tissue phantom, and squeezed a rubber-tipped dropper to draw up and dispense a solution. Anything up to its 15 mm internal diameter, in fact.

Where the Spring Gives Out

It is also, for now, fragile, which is a familiar weakness in fabric pneumatic actuators. In cyclic testing the coil spring fractured after more than 7,000 test runs, and the height it could reach with a load had been sagging since about 5,000; the air pouch itself never leaked, holding tight past 10,000 inflations, and torn seams could be repaired by pressing them together with heat again.

The press release is a shade bolder than the paper. It says the machine can “reliably operate beyond the level of a human hand,” where the paper claims exceedance only in particular measures and then states that the robot is meant to complement dexterous hands rather than replace them; the release also leaves out the fractured spring, and the fact that the aeroengine inspection ran on a 3D-printed mock-up rather than an engine.

Fair enough as a caveat, because the cost argument is the compelling bit anyway. The authors price their robot at about $5 in materials (their own tally, from a parts list in the supplement), against $500 to $50,000 for the dexterous hands it might stand in for. At 9 grams it adds almost nothing to the end of a robot arm, and for a humanoid that matters, since every gram out at the wrist costs stability and battery. Cheap enough to treat as consumable, in other words, which is a different proposition from a hand you insure.

Whether any of it survives contact with a factory is another question; the thing is still tethered to a compressor, and the authors reckon a portable version needs compact micropumps, onboard power and a wireless link before it goes anywhere useful. Still, there is something bracing about a gripper that costs less than lunch and can post itself through a gap barely wider than its own body.

Frequently Asked Questions