Machine learning gives prosthetic limbs “hands”

Researchers are using machine learning, cameras and sensors to improve how to predict the grip strength required for a prosthetic hand.

From the Journal: Nanotechnology and Precision Engineering

A pressure sensor is installed at the fingertip of the prosthetic hand, and a camera is installed near the palm (left). The prosthetic hand was tested with cans, eggs, and USB sticks (right). Credit: Li et al.

WASHINGTON, Jan. 20, 2026 — Holding an egg requires a gentle touch. If you squeeze too hard, it will get dirty. On the other hand, opening a water bottle requires a little more grip strength.

According to the Centers for Disease Control and Prevention, approximately 50,000 new amputations occur in the United States each year. Losing a hand can be especially debilitating and affect a patient’s ability to perform standard daily tasks. One of the main challenges of the prosthetic hand is being able to properly adjust the appropriate grip based on the object being handled.

In “Nanotechnology and Precision Engineering” published by AIP Publishing, researchers from China’s Guilin University of Electronic Technology have developed an object identification system for prosthetic hands to guide appropriate grip force decisions in real time.

“We want to free users from thinking about control methods. [an object] And we let them focus on what they want to do, creating a truly natural and intuitive interaction,” said author Hua Li.

More than 90% of the types of items that patients with disabilities use on a daily basis are fragile items such as pens, cups and bottles, balls, metal sheets such as keys, and eggs. The researchers measured the grip strength required to manipulate these common items and fed these measurements into a machine learning-based object identification system that uses a small camera placed near the palm of the prosthetic hand.

Their system uses an electromyography (EMG) sensor on the user’s forearm to determine what the user is trying to do with an object at hand.

“While electromyographic signals can clearly convey grasp intent, it is difficult to answer the important question of how much force is required, which often requires complex training and adjustments by the user,” said Lee. “Our approach was to leave that ‘how much’ question up to the visual system.”

The group plans to integrate haptic feedback into the system to provide users with an intuitive physical sensation, and additional EMG signals can be used to establish a two-way communication bridge between the user and the hand.

“What we are most excited about and currently focused on is enabling prosthetic hand users to seamlessly and reliably perform fine motor tasks of daily living,” said Lee. “We hope users will be able to effortlessly tie their shoes or button their shirts, confidently lift an egg or glass of water without consciously calculating their strength, peel a fruit naturally, or pass a plate to a family member.”

###