Researchers at Florida Atlantic University have developed a prosthetic hand control system that uses custom 3D-printed sensors and individualized artificial intelligence models to adapt to each user’s muscle patterns.
The work, led by Erik Engeberg, Ph.D., a professor in FAU’s College of Engineering and Computer Science, aims to address one of the main challenges with current prosthetic hands: the difficulty of translating unstable muscle signals into natural, reliable movement. Muscle signals can shift because of sweat, skin changes or everyday motion, making control less intuitive for users.
FAU’s system begins with a 3D scan of a person’s residual limb. Researchers then create a custom 3D-printed wearable sleeve fitted with soft, flexible magnetic sensors. These sensors rest against the skin and capture small changes in muscle shape and pressure as the user attempts hand and wrist movements.
The sensor layout is tailored to each user, with arrays of either 18 or 24 modules depending on limb size and anatomy. The system is then paired with an individualized AI model that learns the person’s own muscle patterns rather than relying on a generalized dataset.
In testing with 10 participants, including three upper-limb amputees, the system classified 19 hand and wrist gestures in real time and translated those signals into control of a dexterous robotic hand.
To test durability, the researchers applied more than 7,500 robotic force cycles over several hours while measuring sensor response. FAU said the system maintained stable signals, accurate pressure capture and repeatable performance, with no meaningful drift or degradation.
“Prosthetic control is not one-size-fits-all. Every individual brings a distinct movement signature shaped by their anatomy, injury history and how their remaining muscles function,” said Engeberg, senior author of the study.
He added that combining 3D-printed wearable sensors with individualized AI models brings prosthetic systems closer to responding “naturally and in real time to a person’s intent.”
The findings also showed that no single sensor setup worked best for every user. Some participants performed better with fewer sensors, while others required more. In several cases, users achieved more than 90% accuracy across multiple gestures only when the sensor layout was tailored to their residual muscles.
Article & Image source: Florida Atlantic University
