View in EDS

Koopman-Based Prediction of Hand Grip Force Using sEMG - Demonstration & Outlook

Saved in:
Bibliographic Details
Title: Koopman-Based Prediction of Hand Grip Force Using sEMG - Demonstration & Outlook
Authors: Kamenar, Ervin, Bazina, Tomislav, Mezić, Igor
Publisher Information: 2025.
Publication Year: 2025
Subject Terms: electromyography, Koopman operator theory, robotic rehabilitation, grip force estimation
Description: Restoring hand function after stroke or musculoskeletal injury requires rehabilitation devices that can sense intention and deliver just-enough assistance. We demonstrate a real-time, ROS-based framework that merges surface electromyography (sEMG) with Koopman operator theory to estimate and forecast hand-grip force for adaptive, assist-as-needed control. A single forearm electrode pair is sufficient: after a one-time calibration of 20–30 seconds, the pipeline filters the sEMG, embeds it in a high-dimensional space, and learns a Koopman model that estimates grip force with high accuracy. The short-term prediction model is updated in real-time via Dynamic Mode Decomposition (DMD), enabling the robot to anticipate user effort and adapt its assistance accordingly. The workshop will include a live demonstration highlighting the system’s accuracy and robustness in grasping tasks. Future work will (i) complete Koopman estimation and prediction system integration, (ii) unify the optimization and sensitivity pipeline for rapid generalization to additional grasps, and (iii) couple data-driven DMD with state-of-the-art sEMG decomposition to link hand dynamics to motor-unit activity—paving the way toward an integrated, novel rehabilitation glove.
Document Type: Conference object
Accession Number: edsair.dris...01492..c641d89dbac5e578e3b1f78dc30879f3
Database: OpenAIRE
Description
Abstract:Restoring hand function after stroke or musculoskeletal injury requires rehabilitation devices that can sense intention and deliver just-enough assistance. We demonstrate a real-time, ROS-based framework that merges surface electromyography (sEMG) with Koopman operator theory to estimate and forecast hand-grip force for adaptive, assist-as-needed control. A single forearm electrode pair is sufficient: after a one-time calibration of 20–30 seconds, the pipeline filters the sEMG, embeds it in a high-dimensional space, and learns a Koopman model that estimates grip force with high accuracy. The short-term prediction model is updated in real-time via Dynamic Mode Decomposition (DMD), enabling the robot to anticipate user effort and adapt its assistance accordingly. The workshop will include a live demonstration highlighting the system’s accuracy and robustness in grasping tasks. Future work will (i) complete Koopman estimation and prediction system integration, (ii) unify the optimization and sensitivity pipeline for rapid generalization to additional grasps, and (iii) couple data-driven DMD with state-of-the-art sEMG decomposition to link hand dynamics to motor-unit activity—paving the way toward an integrated, novel rehabilitation glove.