Related Experiment Video
Updated: Sep 23, 2025
11:18
Quantifying Learning in Young Infants: Tracking Leg Actions During a Discovery-learning Task
Published on: June 1, 2015
10.8K
Multi-Day EMG-Based Knee Joint Torque Estimation Using Hybrid Neuromusculoskeletal Modelling and Convolutional Neural
Robert V Schulte1,2, Marijke Zondag1,2, Jaap H Buurke1,2
1Roessingh Research and Development, Enschede, Netherlands.
Frontiers in Robotics and AI
|May 13, 2022
Summary
Convolutional neural networks (CNNs) best estimate knee torque for transfemoral prosthesis control over multiple days. While CNNs show lower error rates, neuromusculoskeletal models offer day-to-day robustness.
Area of Science:
- Biomedical Engineering
- Rehabilitation Robotics
- Signal Processing
Background:
- Surface electromyography (EMG) offers intuitive control for transfemoral prostheses but is susceptible to noise and variability.
- Accurate multi-day knee torque estimation is crucial for reliable prosthetic control, yet optimal modeling approaches remain unclear.
Purpose of the Study:
- To compare the suitability of three modeling frameworks for estimating knee torque using EMG over multiple days.
- To evaluate the performance and robustness of Convolutional Neural Network (CNN), Neuromusculoskeletal (NMS), and Hybrid models for non-weight-bearing prosthetic control.
Main Methods:
- Three models were developed: a direct CNN mapping EMG to knee torque, an NMS model using EMG and biomechanical parameters, and a Hybrid model combining CNN for muscle activation with NMS components.
- Multi-day non-weight-bearing measurements were collected from ten able-bodied participants.
Main Results:
- The direct CNN model achieved the lowest error (NRMSE 9.2 ± 4.4%), outperforming both Hybrid (12.4 ± 3.4%) and NMS (14.3 ± 4.2%) models.
- CNN and Hybrid models showed significant performance degradation after the first day, unlike the NMS model which remained robust across measurement days.
Conclusions:
- CNNs demonstrate superior accuracy for multi-day knee torque estimation in prosthetic applications.
- While CNNs excel in error reduction, NMS models provide essential day-to-day robustness, highlighting a trade-off for prosthetic control system design.
