Researchers at the Technical University of Munich have developed a soft-hand exoskeleton glove that restores grasping ability for people with hand paralysis, using electrical signals read directly from forearm muscles combined with a machine-learning model that detects a user’s intended movement with 97% accuracy. Rather than requiring implanted electrodes or invasive sensing, the system reads surface muscle activity that remains even when the resulting movement itself doesn’t happen — the same category of signal, notably, that wearables like Wearable Devices’ Mudra Link use for gesture control in non-medical contexts.
Reading intent from muscles that can’t quite move the hand
Paralysis from stroke, spinal cord injury, or nerve damage frequently leaves some residual electrical activity in forearm muscles even when that activity no longer translates into functional finger or wrist movement. TU Munich’s system captures that electrical signal and feeds it into a machine-learning model trained to classify it as a specific intended movement — a grasp, a release, a particular finger configuration — with reported accuracy of 97%, high enough to make the difference between a device that assists real intent and one that fires on noise.
Soft actuation instead of rigid motors
Once intent is classified, the glove responds by inflating soft air cushions built into its structure, physically assisting finger and wrist movement to complete the grasp the user’s muscles were attempting. That soft-pneumatic approach is a deliberate departure from the rigid, motor-driven exoskeletons that have dominated earlier assistive-hand research — air cushions conform to the hand’s actual shape and movement path rather than forcing it through a fixed mechanical linkage, which matters both for comfort during extended wear and for reducing the risk of the device fighting against a hand’s natural range of motion.
Part of a broader wave of muscle-signal wearables
This work lands alongside a broader research trend covered on this site — wearables that read residual electrical or neural signal before or instead of physical movement, whether for restoring lost function (as here) or for augmenting able-bodied gesture control entirely. The throughline across both is the same technical bet: surface-level bioelectric signal carries more usable information than most interfaces have historically been built to extract, and better classification models are what’s finally making that signal reliable enough to act on in real time.
Related Reading
- Exoskeleton Reads Faint Muscle Signals to Overcome Hand Paralysis — Neuroscience News
- Smart exoskeleton glove restores grip for paralyzed hands — Interesting Engineering
- Fabric-Based Wearable Robotic Exoskeleton Gloves: Advancements and Challenges — Advanced Materials Technologies
- KinesCeTI: A Modular and Size-Adaptable Force Feedback Glove with Interchangeable Actuation — arXiv