New EMG method predicts grip force for robotic hand rehabilitation with minimal sensors

Scientists have developed a technique to estimate and forecast grip force using just a single pair of muscle sensors, potentially simplifying the way robotic rehabilitation devices adapt to patients' needs.

The Challenge#

Hand paralysis from stroke, multiple sclerosis, and similar conditions severely limits independence in everyday tasks. While robotic rehabilitation systems can help restore function, they work best when they respond precisely to each patient's actual muscle output. Traditional approaches require many sensors to achieve accurate readings, making devices bulky and complex.

The Approach#

Researchers collected muscle electrical signals (surface electromyography, or sEMG) from 13 participants while they performed medium-strength grasping motions at two different forearm positions. They paired these measurements with a hand dynamometer to verify their results. Using a mathematical framework called Koopman operator theory—a data-driven method for analyzing dynamic systems—they built an algorithm that translates processed sEMG signals into grip force estimates and predictions.

Key Results#

The system achieved remarkable accuracy: it estimated grip force with an error of roughly 5.5%, and predicted grip strength half a second into the future with an error around 17.9%. Importantly, the algorithm proved robust to electrode placement variation—slight shifts in sensor position did not significantly degrade performance. The method also runs extremely fast, processing and predicting a half-second of sEMG data in just 30 milliseconds, making real-time clinical use feasible.

Why It Matters#

By reducing the number of sensors needed while maintaining accuracy, this work could make robotic rehabilitation more practical and accessible. Devices can now better mirror a patient's intent without requiring elaborate sensor arrays, potentially improving outcomes for stroke survivors and those with neuromuscular disease.