Marie D. Schmidt; Tobias Glasmachers; Ioannis Iossifidis
Understanding how movement relates to muscle activation is a core problem in motor control and a practical challenge for rehabilitation technology. In this article, we train a Long Short-Term Memory (LSTM) model on upper-limb end-effector kinematics to predict activity across eight muscles.
The model predicts muscle activity accurately for new repetitions of known movements and also generalizes to previously unseen movements, indicating that it learns structure beyond simple memorization. We further investigate factors that shape this generalization, including movement segmentation, the role of the swivel angle in handling redundancy, and how the diversity/complexity of training movements affects performance.
The study uses an experimental dataset of 23 distinct upper-limb movements performed by five subjects, offering insights relevant to motor neuroscience as well as data-driven human–machine interaction systems.
Reference
Scientific Reports, Nature Portfolio, 2026. ISSN: 2045-2322.
Links
BibTeX
@article{schmidt2026lstm,
title={Insights into Motor Control: Predict Muscle Activity from Upper Limb Kinematics with LSTM Networks},
author={Schmidt, Marie D. and Glasmachers, Tobias and Iossifidis, Ioannis},
journal={Scientific Reports},
year={2026},
publisher={Nature Portfolio},
issn={2045-2322}
}