Related Experiment Video
Updated: Feb 20, 2026
08:48
Combining Multiple Data Acquisition Systems to Study Corticospinal Output and Multi-segment Biomechanics
Published on: January 9, 2016
7.3K
Estimation of joint angle based on surface electromyogram signals recorded at different load levels
Summary
This study trains classification models using surface electromyogram (sEMG) data from various loads to estimate joint angles for upper-limb prostheses. Models trained on diverse loads improved accuracy, though load variations still impacted performance.
Area of Science:
- Biomedical Engineering
- Rehabilitation Robotics
- Signal Processing
Background:
- Surface electromyogram (sEMG) is crucial for controlling upper-limb exoskeletons and prostheses by estimating joint angles.
- Variations in carried load significantly alter sEMG signals, degrading the accuracy of joint angle estimation.
- Developing robust sEMG-based control systems that account for load variability is essential for effective prosthetic and exoskeleton function.
Purpose of the Study:
- To develop and evaluate classification models for joint angle estimation using sEMG data across different carried loads.
- To compare the performance of subject-specific and subject-independent models trained on pooled sEMG data from various loads.
- To assess the impact of load variations on the accuracy of sEMG-based joint angle estimation.
Main Methods:
- sEMG signals were recorded during elbow flexion/extension from three participants under four different load conditions (1-6 Kg) and six joint angles (0-150 degrees).
- Classification models were trained using sEMG data from all loads, evaluated for both subject-specific and subject-independent scenarios.
- Performance was compared against models with explicit knowledge of the carried load.
Main Results:
- The proposed joint angle estimation models, trained on diverse load data and assuming unknown loads, performed significantly above chance level.
- Both subject-specific and subject-independent models demonstrated improved performance when trained on a pool of sEMG data from various loads.
- Transferring models from known to unknown load conditions resulted in a 20% to 32% average accuracy loss in subject-specific classifiers.
Conclusions:
- Training sEMG-based joint angle estimation models with data from multiple load conditions enhances robustness and accuracy.
- While effective, significant accuracy degradation occurs when subject-specific models encounter unencountered load conditions.
- Further research is needed to mitigate the impact of load variability for seamless prosthetic and exoskeleton control.
