Funder: European Research Council under EU Framework Programme for Research and Innovation, Horizon 2020
Collaborators: Otto Bock Healthcare Products GMBH (Austria), OT Bioelettronica (Italy), Aalborg University (Denmark), Technical University of Berlin (Germany), and University of Goettingen (Germany).
In spite of decades of research, advanced man-machine interfaces based on the electromyogram (EMG) still have a significant distance from commercial applications. This is particularly true for myoelectric prosthetic devices. Commercial myoelectric control systems for prostheses can reliably control only one single degree-of-freedom (DOF) at a time. The clinical success of these prostheses is quite limited. Many upper limb amputees are not satisfied with the restoration of quality of life such prostheses can provide. In order to improve the performance and increase the user acceptance of myoelectric prostheses, intuitive and reliable control of multiple DOFs is needed. Academic research has been suggesting advanced signal processing combined with pattern recognition and machine learning to achieve this goal for more than one decade already. And although the research results documented in the literature strongly indicate that such advanced and clinically viable systems are possible, the transfer from academic research to commercial systems has not succeeded yet. AMYO has combined European academic excellence in EMG signal processing and machine learning, industrial expertise in EMG acquisition, and the clinical and industrial expertise of the market leader in prosthetics. The objective of AMYO was to develop myoelectric control systems which are clinically and commercially viable and that provide reliable, intuitive and proportional control of two degrees of freedom.
To achieve that, we analyzed the current academic state-of-the art in myoelectric control, and identified the hurdles for clinical application. This analysis revealed the lack of robustness as the main limiting factor, and the need to improve the three most important elements of the EMG-based man-machine interface: signal acquisition, signal processing, and training of users. Therefore, we developed new signal acquisition methods based on HD-EMG electrodes and silicone based electrodes. We also developed advanced signal processing methods based on neurophysiological modelling and machine learning principles to achieve intuitive, proportional and simultaneous control of two DOF. Furthermore, we developed a novel training system based on performance and psychometric measures. To test our methods for real clinical applicability, we used standardized, state-of-the-art performance evaluation methods and extended them to account for multifunctional prostheses. Our results on intact-limb subjects and subjects with limb deficiency showed considerable improvements compared to the state-of-the art. We are confident that we made a large step closer to a clinically and commercially viable man-machine interface for multifunctional prosthesis. Moreover, the knowledge acquired and the results achieved in this project will help us to extend myoelectric control to other assistive devices such as orthotics, exoskeletal devices, mobility solution, and neurostimulation devices.