3D-printed robot hand stretches up to 7 cm, grips eggs and glass bottles alike

- Joint research by KAIST, KIST and Seoul National University of Science and Technology uses AI to find optimal material formula

- Hand lifts 1 kg water bottle, handles fragile objects safely; applications in soft robotics and wearables

A soft robotic hand made with 3D printing (AI-generated image). [Provided by KAIST]

"Grip a fragile egg gently, and a heavy water bottle firmly."

Researchers have developed a technology to 3D-print a robotic hand that bends like a human finger and adjusts its grip to match the shape and firmness of whatever it holds.

KAIST announced Tuesday that a research team led by Professor Lee Seung-cheol of the mechanical engineering department, working jointly with KIST researcher Na Jong-beom and Seoul National University of Science and Technology Professor Park Beom-su, used AI to develop a material that can be 3D-printed and stretches up to 7 centimeters.

Demand for soft, highly stretchable materials that can be formed into complex shapes has grown as soft robots designed for direct human contact, wearable devices and patient-specific medical devices attract increasing attention.

Achieving both printability and stretchability at the same time, however, had proved difficult.

The team turned to digital light processing, a 3D-printing technique that cures liquid material into a desired shape by exposing it to light. While the method can produce complex structures quickly, increasing a material's stretchability and strength raises its viscosity, making it harder to print. Thinning the material to ease printing, on the other hand, sacrifices both stretchability and strength. Finding a material that prints well and stretches well was the central challenge.

The team used AI to build a dataset linking material formulations to performance. They mixed liquid materials in various combinations, cured them, and measured how far each stretched, how rigid it was, how quickly it hardened under light, and how freely it flowed.

The training data included not only materials that printed successfully but also those too viscous to print at all. Machine learning then mapped the relationship between formulation and performance, and from that model the team identified the optimal combination — one that satisfied both 3D printability and high stretchability.

The AI-identified material printed stably on an actual DLP 3D printer. It proved highly stretchable, extending to more than six times its original length without breaking.

The team used the material to fabricate a soft actuator that moves like a human finger. Soft actuators use air pressure or similar mechanisms to produce smooth, muscle-like motion. When air was pumped in, the device inflated like a balloon and curved naturally, mimicking a bending finger.

The joint research team behind the study. From left: Song Young-han, KIST researcher; Park Beom-su, professor at Seoul National University of Science and Technology; Lee Seung-cheol, KAIST professor; and Na Jong-beom, KIST researcher. [Provided by KAIST]

A soft robotic hand assembled from multiple actuators demonstrated its capabilities in object-gripping tests. It successfully lifted a 1-kilogram water bottle and securely grasped a range of objects varying widely in size, shape and firmness — including fragile eggs, glass bottles, an egg carton and a computer mouse.

"This research is significant in showing that combining experimental data with AI can efficiently identify optimal material combinations that would have been difficult to find through conventional means," Lee said. "We expect it to accelerate the development of 3D-printable materials with the performance characteristics needed across diverse fields, including soft robotics, wearable devices and customized medical devices."

The findings were published in Nature Communications.

nbgkoo@heraldcorp.com