A Robotic Hand That Feels Just Right

This robot hand grips too hard. This robot hand grips too soft. When attempting to replicate a human’s ability to interact with everyday objects of varying textures and materials, existing robot hands tend to fall into one of these two categories. That may change in the future, however, thanks to the work of a group of engineers at Johns Hopkins University. They have developed a novel robotic hand that is just right, so much so that it might even satisfy Goldilocks.

Traditional robotic hands either rely on rigid structures for strong, controlled movements or use soft materials for safer, more flexible interactions. However, neither approach alone can fully replicate the human hand’s ability to grip with the right amount of force while recognizing and adapting to different textures.

The newly developed hybrid robotic hand takes inspiration from human anatomy, combining a rigid internal skeleton with soft, flexible joints. The fingers are constructed with three independently actuated soft robotic joints made from Dragon Skin silicone, a highly elastic and durable material. These soft joints are embedded between rigid, 3D-printed PLA structures, mimicking the way human fingers balance strength and compliance.

Each of the hand’s 14 joints, including three per finger and two for the thumb, moves independently, providing for dexterous and precise movement. Instead of relying on tendons or motors for movement, the fingers use Arduino-controlled pneumatic actuators — air-filled chambers that inflate to create motion. This setup not only enhances the hand’s grasping ability, but also ensures safe interactions with objects and people.

Inspired by the layered mechanoreceptors in human skin, the fingertips contain three distinct sensing layers. The outermost layer mimics Merkel cells and Meissner corpuscles, which detect light touch and low-frequency vibrations. The middle layer replicates Ruffini endings, responsible for detecting deformation. The innermost layer, attached to the rigid fingernail structure, functions like Pacinian corpuscles, sensing high-frequency vibrations and transient pressure changes.

Each of these sensing layers is constructed using different materials and technologies. The outer and middle layers are made of piezoresistive fabric sensors, which change electrical resistance when pressure is applied. The innermost layer uses a piezoelectric transducer, generating voltage in response to force. Together, these three layers enable the robotic hand to differentiate textures and adjust grip strength accordingly.

To process the vast amount of tactile data, the robotic hand employs a neuromorphic encoding system. This approach mimics how the human nervous system transmits information, reducing computational load while maintaining precise and dynamic touch perception. The tactile signals are processed using machine learning algorithms, allowing the robotic hand to recognize different surface textures and adjust its grasp based on real-time feedback.

In laboratory tests, the hybrid robotic hand demonstrated its ability to grasp and manipulate a variety of objects, including delicate items like soft toys and sponges, as well as more rigid items such as metal water bottles and pineapples. It achieved an impressive 99.69% accuracy in distinguishing objects of varying surface textures and compliance. In another texture discrimination task, the robotic hand outperformed traditional soft and rigid robotic fingers, achieving an accuracy of 98.38%.

While still in development, this work could have significant implications not only for prosthetics but also for robotic systems used in fields such as healthcare, manufacturing, and service industries. Future improvements may include stronger grip forces, additional sensors, and more durable materials.