RTAS 2026: From EMG Signal to Grip Force Prediction in 27 ms

May 2026

From EMG signal to grip force prediction in 27 ms — presented at RTAS 2026, CPS-IoT Week, Saint Malo, France 🇫🇷

Tomislav Bazina presented our work on Koopman-driven grip force prediction from sEMG at the 32nd IEEE Real-Time and Embedded Technology and Applications Symposium (RTAS 2026), held as part of CPS-IoT Week in Saint-Malo, France. This work with a particular focus on the real-time timing characterization of the full prediction pipeline is part of our Croatian Science Foundation project and contributes to the project result: “Open-source package, algorithms for real-time processing, filtering, and calibration of muscle activity detected using surface electromyography (sEMG) sensors”, related to Objective O2.

A key insight highlighted at the conference was that the Koopman operator approach is a natural fit for real-time systems. Instead of relying on computationally complex procedures with variable execution cost, the Koopman model performs the same operations at each step, making its runtime predictable and easy to bound.

Our results show that the full pipeline — from raw EMG signal to grip force prediction — executes in at most 27.1 ms, which is comfortably within the 100 ms real-time deadline.

The work was also presented as a poster during the Q&A session, where it attracted interest from the real-time systems community, particularly regarding the use of Koopman methods as a principled approach to predictable, low-latency biomedical signal processing.

Figure: Measured online compute breakdown across the pipeline stages: FFT preprocessing, force estimation, predictor update, and forecast.

For more information, see:

Copyright notice: The article PDF is provided for personal and scholarly reference only. Copyright remains with the original publisher and/or authors as applicable. Conference program materials remain the property of RTAS 2026 / CPS-IoT Week and their respective organizers. Images and presentation materials are used here for non-commercial academic dissemination.