Koopman-driven real-time sEMG signal decomposition for robotic rehabilitation - uniri-iz-25-116 (2025-2029)
Supported by the University of Rijeka
Loss of hand function caused by conditions such as stroke or multiple sclerosis severely limits activities of daily living. Rehabilitation robotics offers effective therapeutic solutions while reducing the need for intensive therapist involvement. Electromyography (EMG)-based robotic rehabilitation provides clear advantages over conventional approaches and open-loop control devices, while soft robotics—drawing inspiration from natural organisms and plants—enables safer and more compliant interaction with the human body.
The proposed research project builds on prior work presented in Koopman-driven grip force prediction through EMG sensing and extends these ideas toward real-time surface electromyography (sEMG) signal decomposition using Koopman operator theory (KOT) and dynamic mode decomposition (DMD). The objective is to identify motor unit (MU) activity and correlate it with grip strength across different grasp types, with a particular focus on real-time implementation. :contentReference[oaicite:0]{index=0}
Koopman operator theory and data-driven DMD provide a powerful framework for decomposing complex EMG dynamics into dominant spatiotemporal components, enabling the extraction of frequency and damping features relevant to neural control. By leveraging these advanced analytical tools, the project seeks to address key limitations of existing methods, particularly in terms of accuracy, interpretability, and real-time applicability.
The project also draws on the principal investigator’s earlier work in soft robotics, including Control of soft robots with inertial dynamics”, which demonstrated the potential of Koopman-based approaches for modeling and control in highly dynamic soft robotic systems. Together, these research directions create a strong foundation for the development of adaptive robotic rehabilitation devices capable of more precise force control and improved responsiveness to patient-specific needs.
Collaborators
Tomislav Bazina
Jelena Srnec Novak
David Liović
Goran Gregov
Igor Mezić (University of California, Santa Barbara, USA)
Project Dissemination
Papers
Modeling Soft Rehabilitation Actuators: Segmented PRB Formulations with FEM-Based Calibration(Bazina et al., 2026)
Conferences
Journal-First: Koopman-Driven Grip Force Prediction Through {EMG} Sensing(Bazina et al., 2026)
From Koopman Soft Robot Control to {EMG}-Informed Rehabilitation Devices(Kamenar, 2026)
3D Positioning of a Stewart Platform Using Soft Pneumatic Actuators: A Design Approach(Šoljić et al., 2025)
Soft pneumatic glove actuators for hand rehabilitation require compact, accurate models that can be evaluated in real time. At the same time, high-fidelity finite element (FE) simulations are too slow for iterative design and control. We develop a finite element-based calibration pipeline that combines a dependency-constrained human finger kinematic model with a segmented pseudo-rigid-body (PRB) description of ribbed-bellow soft pneumatic actuators sized to individual fingers. FE models with symmetry and contact generate pressure–pose data for the MCP, PIP, and DIP spans, from which we extract per-segment bending angles and axial elongations, fit simple pressure–kinematics relations, and identify PRB parameters using basin-hopping global optimization. The calibrated PRB reproduces FE flexion–extension trajectories for index and little finger actuators with millimetric accuracy (mean segment positioning errors of approximately 2.3 mm and 0.7 mm), preserves finger-like bending localized in the bellows, and maintains negligible compression of inter-joint links (below 1.2%). The pressure–bend and pressure–elongation maps achieve near-unity adjusted R^2, and the PRB forward kinematics evaluates complete pressure trajectories in less than half a millisecond, compared with several hours for the corresponding FE simulations. This pipeline provides a practical route from detailed FE models to controller-ready reduced-order surrogates for design-space exploration and patient-specific control of soft rehabilitation actuators.
@article{bazina2026modeling,title={Modeling Soft Rehabilitation Actuators: Segmented PRB Formulations with FEM-Based Calibration},author={Bazina, Tomislav and Liovi{\'c}, David and Srnec Novak, Jelena and Kamenar, Ervin},journal={Actuators},year={2026},volume={15},number={1},pages={22},keywords={soft pneumatic actuator (SPA); finger rehabilitation; pseudo-rigid-body (PRB) model; finite element method (FEM); hand kinematics; reduced-order modeling; patient-specific design},doi={10.3390/act15010022},}
RTAS2026
Journal-First: Koopman-Driven Grip Force Prediction Through EMG Sensing
Tomislav Bazina, Ervin Kamenar, Maria Fonoberova, and Igor Mezić
2026
Extended paper, presentation slides, and poster available online
We study grip force estimation and short-horizon prediction from surface electromyography (sEMG) under real-time constraints for robotic rehabilitation. Using a single pair of sEMG sensors, signals are processed in fixed 0.5 s batches, and Koopman-based models with problem-specific lifting are applied to estimate the current grip force and predict it 0.5 s ahead with low weighted mean absolute percentage error (wMAPE). The timing requirements include a soft setup-time constraint for on-site calibration and model training, together with an online deadline requiring each batch to produce a forecast shortly after its arrival. The system performs one-time calibration, fits the estimation model in approximately 1.5 s, and achieves a maximum observed end-to-end latency of 27.1 ms per batch, covering sEMG preprocessing, force estimation, prediction-model training, and forecasting. These results leave sufficient computational margin for real-time operation. This paper extends the original journal work by introducing an explicit timing specification, clarifying the interpretation of real-time deadlines, and reporting the current status of ROS 2 integration.
@misc{Bazina2026RTASJournalFirst,author={Bazina, Tomislav and Kamenar, Ervin and Fonoberova, Maria and Mezi{\'c}, Igor},title={Journal-First: Koopman-Driven Grip Force Prediction Through {EMG} Sensing},eventtitle={32nd IEEE Real-Time and Embedded Technology and Applications Symposium},eventdate={2026-05-12/2026-05-14},venue={Online},year={2026},type={Journal-First presentation},url={https://ekamenar.github.io/assets/img/news/RTAS2026/Journal_first__Koopman_Driven_Grip_Force_Prediction_Through_EMG_Sensing-3.pdf},related={https://ekamenar.github.io/assets/img/news/RTAS2026/rtas2026_j1_talk_slides.pdf,
https://ekamenar.github.io/assets/img/news/RTAS2026/rtas2026_j1_poster.pdf},note={Extended paper, presentation slides, and poster available online},}
RoboSoft2026
From Koopman Soft Robot Control to EMG-Informed Rehabilitation Devices
@misc{Kamenar2026RoboSoftInvited,author={Kamenar, Ervin},title={From Koopman Soft Robot Control to {EMG}-Informed Rehabilitation Devices},howpublished={Invited talk at the workshop ``The Role for Control in Soft Robot Autonomy,'' IEEE RoboSoft 2026},address={Kanazawa, Japan},date={2026-04-08},year={2026},note={Abstract and presentation},}
2025
ICMIE25
3D Positioning of a Stewart Platform Using Soft Pneumatic Actuators: A Design Approach
Antonio Šoljić, Goran Gregov, and Ervin Kamenar
In Proceedings of the 11th World Congress on Mechanical, Chemical, and Material Engineering (MCM’25) , 2025
In this paper, the application of a novel soft bellows pneumatic actuator (SBPA) into an advanced mechatronic system specifically, a Stewart platform, is investigated. Our previously research has demonstrated that the newly designed SBPA can generate forces exceeding 100 N, achieving a contraction ratio greater than 40% relative to its maximum length, and reaching motion speeds above 60 mm/s. Moreover, precise linear positioning within 10 μm has been achieved through the application of a Linear Quadratic Regulator (LQR). To further evaluate the capabilities of the developed actuator, a six-degree-of-freedom Stewart platform was designed using three identical SBPAs. The 3D positioning of the platform was evaluated under open-loop control, using a camera-based system to track the displacement of key points on the platform for model identification and validation of the control results. The developed Stewart platform achieved a positioning error of 1.1% at the centre of the platform when all SBPAs were activated. Additionally, the platform’s dynamic performance was assessed by actuating the SBPA with sinusoidal inputs at varying frequencies. At 1 Hz, the platform exhibited consistent vibrational motion, indicating its potential for use in vibration-based applications. This study advances the development of SBPAs and provides insight into their integration in complex mechatronic systems.
@inproceedings{vsoljic20253d,title={3D Positioning of a Stewart Platform Using Soft Pneumatic Actuators: A Design Approach},author={{\v{S}}olji{\'c}, Antonio and Gregov, Goran and Kamenar, Ervin},booktitle={Proceedings of the 11th World Congress on Mechanical, Chemical, and Material Engineering (MCM'25)},year={2025},organization={Pariz: International ASET Inc.},}
