Development of real-time neuromechanical models for intuitive prosthetic ankle control across walking conditions
Federica Damonte is a PhD student in the department Neuromechanical Engineering. (Co)Promotors are prof.dr.ir. M. Satori and prof.dr.ir. H. van der Kooij from the faculty Engineering Technology (ET), University of Twente and dr. G.V. Durandau from the McGill University of Quebec.
Powered leg prostheses have been introduced to enhance mobility and reduce energy consumption during walking; however, they often fail to meet user expectations. While one might assume that robotic systems should assist users, at the time of the thesis development the technology was not advanced enough to understand and respond to user needs in a natural way, as a biological leg would. As a consequence of existing technologies, patients adopted non-natural compensatory movements that require higher energy expenditure and result in problems such as trauma disorders, instability, and inefficient walking that affect their quality of life.
Furthermore, the solutions developed were either limited in the movement they could perform or too complex (many parameters to be tuned), hindering a broader range of movement.
State-of-the-art prosthetic devices are controlled by using a simple set of rules to activate predefined changes in the damping and/or actuation of the knee and ankle joints, while the activation of biological joints is based on continuous modulation of impedance and active forces.
Our investigations focused on the development and evaluation of an emerging methodology to control wearable robotics: the use of neuromechanical modeling as a mid-level controller.
A neuromechanical model is a digital model of the body that can simulate the dynamics of muscle force production, including neuro-muscular commands. Compared to pre-existing prosthetic controllers it would enable mimicking the biomechanical function of an intact limb decoding plausible torque commands to the prosthetic according to the subject-specific requirements across various movement conditions.
We investigated two approaches to generate feedforward control inputs to the model: from electromyography (EMG) signals, to improve voluntary control from the user allowing for non-steady movements, that are not accounted for by other models, and from the synergy model which can estimate the activity of leg muscles for different walking speeds and elevations, facilitating walking in various conditions with minimal sensor dependency and cognitive effort.
Moreover, our study aimed to develop a solution that would fit different populations of users. Throughout our research, we recruited three subjects with transtibial amputation, two subjects underwent an agonist-antagonist myoneural interface (AMI) surgical procedure [1] while the third subject had a bone-anchored prosthesis (BAP) using an osseointegration implant (OI).
In the first part of the research, we developed a human prosthesis interface using an EMG-driven neuromechanical model that includes a musculoskeletal model, which serves as a detailed digital representation of a participant's intact limb. This enabled the efficient decoding of mechanical torques generated in the participant's missing ankle joint, which could be used to command prosthetic joint rotations directly.
The advantage of EMG control is that by interfacing directly with the user, EMG signal measurements enable versatile prosthesis control that is able to adapt to various situations.
Based on previous literature studies [2], we hypothesized that training individuals with transtibial amputations in learning how to generate biomechanically valid EMG signals can be beneficial for both AMI and BAP users. Moreover, the EMG-driven approach would allow for the continuous EMG decoding of biomechanically plausible joint torques, which could enable the powered prosthesis to closely mimic the biomechanical function of intact limbs not only during overground walking but also on inclines.
However, individuals with transtibial amputation develop different walking strategies than able-bodied individuals, depending on the type, the level of amputation, on the type of prosthesis prescribed. In terms of muscle activity, residual muscles co-contract for stabilizing the prosthetic socket, and it is particularly impactful in reducing the effectiveness of EMG control.
Since the quality of the EMGs is highly variable, such a solution is not adequate for all prosthetic users.
Moreover, volitional EMG control of powered leg prostheses may not always be required. During prolonged motor activities, such as walking long distances, continuous volitional control might lead to both muscle and cognitive fatigue, potentially compromising signal integrity and bionic limb control performance [3][4].
In light of the outlined issues, it is important to explore and pursue an alternative solution that does not fully rely on EMG recordings.
The second part of the project was dedicated to the evaluation of the output of the neuromechanical model combined with a surrogate model of the EMG signals, the synergy model, as it was demonstrated that the muscle activation signals of the leg muscles during human adult locomotion can be reconstructed as a linear combination of four basic patterns [5][6].
In our framework, the EMG signal input was substituted by a new software extension dedicated to the computation of muscle excitations in real-time from a synergy model.
After evaluating the outputs of the synergy-driven model offline, we developed the second human prosthesis interface to control an ankle prosthesis using the same basic framework as in the first case.
We assumed that the main advantage of the synergy-driven approach is that, once a set of basic patterns is defined, no EMG recordings would be needed for the model simulation since they can be replaced by the synergy model, facilitating walking in various conditions with minimal sensor dependency. The framework would be able to generate accurate biomimetic joint torques, and it would benefit the time response of the controller to changes in speed compared to existing neuromechanical models [7]. Future work should aim to combine both approaches to enable volitional control when required and an automated prosthetic behavior during cyclical steady-state movements.
Throughout the project, we developed a real-time controller based on neuromechanical models to simulate the movements of people with transtibial amputation phantom legs using sensory information. This controller allows for real-time actuation of an ankle prosthetic in the torque domain.
Furthermore, we have developed a precise mechanistic model of the lower limb utilizing musculoskeletal models that do not necessitate electrophysiological input data but instead utilize EMG surrogates.
References
[1] Berger, L., Sullivan, C.L., Landry, T., Sparling, T.L. and Carty, M.J., 2023. The ewing amputation: Operative technique and perioperative care. Orthoplastic Surgery, 13(C), pp.1-9.
[2] Huang, S., Wensman, J.P. and Ferris, D.P., 2015. Locomotor adaptation by transtibial amputees walking with an experimental powered prosthesis under continuous myoelectric control. IEEE Transactions on Neural Systems and Rehabilitation Engineering, 24(5), pp.573-581.
[3] Huang, S. and Huang, H., 2018. Voluntary control of residual antagonistic muscles in transtibial amputees: reciprocal activation, coactivation, and implications for direct neural control of powered lower limb prostheses. IEEE Transactions on Neural Systems and Rehabilitation Engineering, 27(1), pp.85-95.
[4] Fleming, A., Stafford, N., Huang, S., Hu, X., Ferris, D.P. and Huang, H.H., 2021. Myoelectric control of robotic lower limb prostheses: a review of electromyography interfaces, control paradigms, challenges and future directions. Journal of neural engineering, 18(4), p.041004.
[5] Lacquaniti, F., Ivanenko, Y.P. and Zago, M., 2012. Patterned control of human locomotion. The Journal of physiology, 590(10), pp.2189-2199.
[6] Gonzalez-Vargas, J., Sartori, M., Dosen, S., Torricelli, D., Pons, J.L. and Farina, D., 2015. A predictive model of muscle excitations based on muscle modularity for a large repertoire of human locomotion conditions. Frontiers in computational neuroscience, 9, p.114.
[7] Dzeladini, F., Van Den Kieboom, J. and Ijspeert, A., 2014. The contribution of a central pattern generator in a reflex-based neuromuscular model. Frontiers in human neuroscience, 8, p.371.
