Electromyography data for non-invasive naturally-controlled robotic hand prostheses

Manfredo Atzori, Arjan Gijsberts, Claudio Castellini, Barbara Caputo, Anne-Gabrielle Mittaz Hager, Simone Elsig, Giorgio Giatsidis, Franco Bassetto, Henning MüllerScientific 2014 · DOI

Abstract

Recent advances in rehabilitation robotics suggest that it may be possible for hand-amputated subjects to recover at least a significant part of the lost hand functionality. The control of robotic prosthetic hands using non-invasive techniques is still a challenge in real life: myoelectric prostheses give limited control capabilities, the control is often unnatural and must be learned through long training times. Meanwhile, scientific literature results are promising but they are still far from fulfilling real-life needs. This work aims to close this gap by allowing worldwide research groups to develop and test movement recognition and force control algorithms on a benchmark scientific database. The database is targeted at studying the relationship between surface electromyography, hand kinematics and hand forces, with the final goal of developing non-invasive, naturally controlled, robotic hand prostheses. The validation section verifies that the data are similar to data acquired in real-life conditions, and that recognition of different hand tasks by applying state-of-the-art signal features and machine-learning algorithms is possible.

Introduction: The Reality Gap in Prosthetics

While robotic hand hardware has become increasingly sophisticated, the control systems governing them remain a significant hurdle for users. Most current myoelectric prostheses rely on simple "open/close" triggers that feel unnatural and require extensive training. This "reality gap" between laboratory success and real-world utility stems from several factors: small and non-standardized datasets, the difficulty of translating results from intact subjects to amputees, and a lack of common benchmarks for comparing algorithm performance.

The Ninapro Database

To address these challenges, the Ninapro (Non-Invasive Adaptive Prosthetics) database was created as a comprehensive, open-access resource for the research community. It provides high-resolution data from 78 subjects, including 11 trans-radial amputees, across three sub-databases:

  • DB1: 27 intact subjects performing 52 movements (10 electrodes).
  • DB2: 40 intact subjects performing 50 movements (12 electrodes, high frequency).
  • DB3: 11 trans-radial amputated subjects performing 50 movements.

Methodology: Multi-modal Data Acquisition

The researchers employed a multi-modal sensor setup to capture the complex relationship between neural intent and physical action. Muscular activity was recorded using surface electromyography (sEMG) electrodes placed on the forearm. For intact subjects, hand kinematics were tracked using a 22-sensor CyberGlove II, while grasp force was measured via a custom Finger-Force Linear Sensor (FFLS).

Overview of the movement replication, classification, and robotic control pipeline.

A key technical contribution is the implementation of a relabelling algorithm. By using a generalized likelihood ratio, the authors aligned the movement labels with the actual sEMG bursts, correcting for human reaction time variability and improving the quality of the training data.

Key Insights and Results

Validation of the database using state-of-the-art classifiers like Random Forests and Support Vector Machines (SVM) demonstrated the feasibility of recognizing complex hand tasks. For intact subjects, an average accuracy of 75.32% was achieved across 50 distinct movements.

However, the study revealed a significant "amputee performance gap," with classification accuracy dropping to approximately 46% for amputated subjects. This highlights the impact of physiological changes such as muscle atrophy and cortical reorganization. The data also suggests that clinical factors, including the amount of remaining forearm muscle and the presence of vivid phantom limb sensations, are strong predictors of successful signal decoding.

Limitations and Future Directions

Despite the promise of the Ninapro database, several challenges remain. For amputees, "ground truth" labels are based on intent rather than physical measurement, introducing potential noise. Additionally, factors like muscle fatigue and the need for real-time, low-latency control (<200ms) are critical for making these systems viable in daily life. Future research will focus on "closing the loop" to ensure robotic prostheses function as seamless extensions of the human body.

Summary updated

Abstract

Recent advances in rehabilitation robotics suggest that it may be possible for hand-amputated subjects to recover at least a significant part of the lost hand functionality. The control of robotic prosthetic hands using non-invasive techniques is still a challenge in real life: myoelectric prostheses give limited control capabilities, the control is often unnatural and must be learned through long training times. Meanwhile, scientific literature results are promising but they are still far from fulfilling real-life needs. This work aims to close this gap by allowing worldwide research groups to develop and test movement recognition and force control algorithms on a benchmark scientific database. The database is targeted at studying the relationship between surface electromyography, hand kinematics and hand forces, with the final goal of developing non-invasive, naturally controlled, robotic hand prostheses. The validation section verifies that the data are similar to data acquired in real-life conditions, and that recognition of different hand tasks by applying state-of-the-art signal features and machine-learning algorithms is possible.
BibTeX
@inproceedings{atzori2014electromyography,
  author = {Manfredo Atzori and Arjan Gijsberts and Claudio Castellini and Barbara Caputo and Anne-Gabrielle Mittaz Hager and Simone Elsig and Giorgio Giatsidis and Franco Bassetto and Henning Müller},
  title = {{Electromyography data for non-invasive naturally-controlled robotic hand prostheses}},
  booktitle = {Scientific},
  year = {2014},
  doi = {10.1038/sdata.2014.53},
}
Figure