In the field of high-end equipment manufacturing, innovations in prosthetic technology are bringing new hope to people with limb loss. Losing a limb due to injury, accident, or disease severely impacts quality of life and makes everyday activities difficult. Recent technological advances, however, are paving the way for more comfortable, intelligent, and intuitive prosthetics that allow users to perform a wider range of tasks with ease.
Over the past decade, many smart prosthetics have been operated using myoelectric signals — electrical impulses generated by muscles and detected by sensors on the wearer's skin. While some systems have shown impressive results, they require users to consciously generate specific muscle signals for each movement, placing considerable physical and mental demand on the wearer.
Researchers at Memorial University of Newfoundland, Canada, have now achieved a breakthrough by developing a new automated prosthetic hand motion control method that does not rely on myoelectric or any other biological signals. An overview of the control system was published in a paper on the arXiv preprint server. The system, based on machine learning models, is trained using video clips of a prosthetic hand performing specific tasks and can autonomously plan and execute the movements required to complete those tasks.
Senior author Professor Xianta Jiang told Tech Xplore that the inspiration for the system came from the desire to make prosthetic hands easier to use. Traditional systems that rely on muscle signals are not only difficult to control but also cause user fatigue. The team wanted to explore whether an autonomous system could take over part of the workload — much like a robot that can "see" and "feel" the world.
Professor Jiang and his colleagues aimed to create a prosthetic hand that can autonomously handle its surroundings and perform grasping tasks with minimal effort from the wearer. Instead of relying on biological signals or explicit user commands to plan movements, the control system uses data from a small camera mounted on the prosthetic wrist combined with sensors that detect touch and motion.
First author Kaijie Shi explained that these inputs are integrated through artificial intelligence (AI) using imitation learning techniques. The AI model learns from past demonstrations — observing how objects are picked up, held, and released — and the prosthetic hand then applies that knowledge to make real-time decisions. The system's unique advantage is that it does not depend on muscle signals; instead, it "understands" objects and tasks, making operation far more natural and intuitive for the user.
To test the new control system, the researchers deployed it on a real prosthetic hand and conducted a series of real-world experiments. Results showed that even when trained with just a few videos of the same person handling a limited set of objects, the system enabled the prosthetic hand to successfully grasp target items with a success rate exceeding 95%.
Professor Jiang said this represents an important step toward creating prosthetics that can work autonomously and reliably in everyday environments. In the future, prosthetic users could benefit from a device capable of performing common tasks — such as picking up a cup or opening a door — without having to consciously think about every single movement.
The researchers plan to continue refining the imitation-learning-based approach and test it in broader experiments involving more individuals who could benefit from advanced prosthetic systems. Ultimately, they hope the system will drive progress in commercial prosthetic hands, reducing the effort required to operate them.
Professor Jiang added that the next step is to test the system with actual prosthetic users and gather feedback to improve its adaptability to different environments and more complex tasks, such as handling soft or irregularly shaped objects. The team also plans to explore applying this technology to other assistive devices, such as exoskeletons for stroke rehabilitation.
