Researchers at the Technical University of Munich (TUM) have developed a soft robotic glove that enables people with paralyzed hands to grasp and manipulate everyday objects. By combining wearable robotics with artificial intelligence, the prototype detects a user's intention to move and translates it into assisted hand movements. The technology could eventually support people whose hand function has been impaired by neurological disorders such as amyotrophic lateral sclerosis (ALS), stroke or peripheral nerve injuries.
Unlike rigid robotic exoskeletons, the newly developed "soft-hand exoskeleton" is built around a lightweight fabric glove fitted with inflatable air chambers. These pneumatic cushions provide targeted assistance to individual fingers and the wrist, allowing users to perform functional tasks such as holding cutlery, grasping a drinking glass or carrying a plate. The system is designed to restore natural hand movements while remaining lightweight, flexible and comfortable enough for everyday use.
AI interprets the user's intention
A central challenge in assistive robotics is determining exactly when a user intends to move. Rather than relying on manual controls, the TUM researchers developed an AI-driven control system that interprets electrical muscle activity in the forearm.Surface sensors continuously record electromyography (EMG) signals generated by muscles that remain partially functional. Machine-learning algorithms analyze these signals in real time to predict grasping intentions with an accuracy of approximately 97 percent.
Once the system detects an intended movement, compressed air inflates the glove's air chambers through 13 separate pneumatic channels. Each finger can flex or extend individually, while the wrist can also rotate, allowing users to perform coordinated grasping actions.To improve safety during object handling, the glove incorporates additional motion sensors that recognize transport movements. These sensors ensure that once an object has been grasped, the glove maintains sufficient grip strength until the movement has been completed, reducing the risk of accidentally dropping objects. According to researcher Nicolas Berberich, this additional motion analysis enables the exoskeleton to keep objects securely in the hand throughout daily activities.
Low-cost with real-world potential
Beyond its intelligent control system, affordability was a key design objective. The glove itself is made from inexpensive textile materials and was sewn by the research team. Despite its relatively simple appearance, the prototype delivers sophisticated functionality at a fraction of the expected cost of many robotic rehabilitation systems. "Our solution is intelligent in two ways," explained researcher Dr. John Nassour. "We've developed a highly reliable method for predicting grasping movements while creating hardware that optimally supports those intended movements."
Professor Gordon Cheng, director of the Institute for Cognitive Systems, emphasized that keeping production costs low could make the technology accessible to a much larger patient population. He also highlighted the importance of working closely with end users throughout the development process. One patient with ALS played a particularly important role during testing. Although he had already lost most hand function, he retained limited movement of the first thumb joint. Researchers used the remaining electrical activity from the flexor pollicis longus muscle as the control signal for the glove.
First successful tests in ALS patient
Despite the patient's extremely weak muscle signals, the AI system correctly identified grasping intentions in nine out of ten attempts. After only five minutes of training, the participant was able to perform several tasks that had previously become impossible. He successfully grasped everyday objects, transferred small cubes into a container and, for the first time in four years, held a fork independently. Researchers also incorporated a video game in which the patient controlled a character's jump using only thumb movements, helping improve control of the interface.
The team is now adapting the technology for additional patient groups, including stroke survivors and people with flaccid paralysis caused by peripheral nerve damage or polyneuropathy. According to neurologist Tobias Wächter of Klinik Passauer Wolf, the glove has the potential to support a broad range of neurological conditions that impair hand function. While larger clinical studies will be needed before routine implementation, the proof-of-concept demonstrates that combining soft robotics with AI-driven intention detection can restore meaningful grasping ability, even in patients with severe motor impairments.
AI-exoskeleton
Last year, Georgia Tech researchers developed an AI-powered robotic hip exoskeleton that automatically adapts to the walking pattern of stroke survivors, helping them walk more efficiently and with less effort. Unlike conventional exoskeletons, which require extensive manual calibration, the new system uses neural networks to learn an individual's gait within one to two minutes. Sensors continuously monitor each step, allowing the AI to adjust support in real time as walking patterns change.
In tests, the technology reduced gait-tracking errors by 70 percent compared with conventional systems, enabling users to walk farther while experiencing less fatigue. The researchers also developed software that works across different exoskeleton platforms, reducing error rates by more than 75 percent after only ten calibration steps. The study, published in IEEE Transactions on Robotics, highlights the potential of AI to deliver more personalized and accessible robotic rehabilitation after stroke.
