Tesi etd-01142020-101701
Link copiato negli appunti
Tipo di tesi
Dottorato
Autore
RODRIGUEZ MARTINEZ, ITZEL JARED
URN
etd-01142020-101701
Titolo
Investigation of the grasp force from surface EMG during grasping and manipulation
Settore scientifico disciplinare
Istituto di Biorobotica
Corso di studi
Istituto di Biorobotica - BIOROBOTICS
Commissione
relatore Dott. CIPRIANI, CHRISTIAN
Parole chiave
- amputees
- emg
- grasp force
- regression
- upper limb
Data inizio appello
03/04/2020;
Disponibilità
completa
Riassunto analitico
Grasping and manipulating objects are tasks so natural for us, that we perform them all the time without even considering all the factors that influence them, like position of the joints, force or speed.
However, individuals who have incurred in a hand loss, may find these tasks to become hard, awkward or unnatural, mostly due to the current state of the art of the prostheses. Myoprosthesis are devices that transform electromyogram (EMG) signals into commands for a powered, electric prosthesis. Although they have significantly improved since the first prototypes back in 1948, they cannot fully replicate yet, the natural movements of a sound limb. In order to improve the capacities of the myoprostheses, a considerable amount of research has been carried out for the decodification of the movement of intent. Nevertheless, an essential factor, the grasp force, has been relegated to a secondary plane.
Regarding this factor, the current myoprostheses approach is to set it proportional to the amplitude of the forearm EMG signals. This representation does not fully correspond to the real relationship between force and EMG and therefore, the force exerted by the myoprostheses is only an approximation of the force a sound limb would produce. Recently, commercial myoprosthesis have started to address this issue by implementing adaptive strategies for the control of the force. However, these strategies aim at preventing the objects from slipping, not actually at reproducing the behavior of a sound limb by decoding the grasp force from the EMG.
Several studies, following a number of methodologies, have been performed to estimate the force from EMG signals, achieving notable accuracies. Nevertheless, all of them focus either on a continuous estimation of the force, or on estimating it when both the force and the EMG signals have reached a stable level. These approaches are difficult to implement in a real-time system, particularly a myoprosthesis, which must comply with strict processing times in order for the action to be felt as natural by the user.
To improve the myoprosthesis response times, research has recently taken advantage of the preplanned nature of grasping movements in humans. This means that the information about the movement should be available before the movement itself. To seize this information, research has recently decanted from targeting the stable phase of the EMG (steady state) to target the transient phase for movement classification. The transient phase, i.e. from the start of the contraction until the signal reaches a stable level, has the advantage of allowing a faster response from the controller. Moreover, due to its temporal structure, the transient hints to be informative about the movement. However, this information has not been exploited for force estimation.
The objective of this dissertation is “the investigation of the grasp force from surface EMG during grasping and manipulation”. This objective was achieved in two steps. First we proved with offline HD-EMG recordings from able-bodied participants that the transient EMG contains information about the target grasp force. The last one referring to the stable plateau value of the force exerted by the subject during the movement. In order to achieve the best estimation possible, several tests were made to determine the fittest algorithm and parameters to do so. By means of this approach we made possible a faster estimation of the movement’s force of intent than with traditional continuous methods.
Finally, we integrated the methodology, algorithm and parameters found in the first stage, to develop a platform to determine if the information contained on the transient about the grasp force could be extracted in a real-time system. This platform was tested both with able-bodied participants and amputees. With that system, we were able to estimate the final GF from the transient EMG with high accuracy. On top of that, the results from the residual limb of the amputees were equivalent to the ones of the non-amputees. This paves the way towards controllers of myoprostheses that are faster and more akin to human behavior.
We also contributed to the acquisition and publication of another database of HD-EMG recordings and hand kinematics. This database could be used to test not only the classic methods of control but also the extraction of the neural drive (high sample frequency of the EMG), performing electrode selection (HD-EMG data) and the development of proportional methods of control (hand kinematics data).
Overall, we increased our knowledge about the physiological relationship under investigation: EMG and force, a rapport that is often oversimplified in the myoprostheses field. Our approach of estimating the grasp force from the transient EMG sheds new light on the relationship of both elements and on the information contained on the transient. Finally, our methodology indicates a faster way to obtain an estimate of the force of intent that can be exploited to develop robust and natural strategies of control for myoprostheses.
However, individuals who have incurred in a hand loss, may find these tasks to become hard, awkward or unnatural, mostly due to the current state of the art of the prostheses. Myoprosthesis are devices that transform electromyogram (EMG) signals into commands for a powered, electric prosthesis. Although they have significantly improved since the first prototypes back in 1948, they cannot fully replicate yet, the natural movements of a sound limb. In order to improve the capacities of the myoprostheses, a considerable amount of research has been carried out for the decodification of the movement of intent. Nevertheless, an essential factor, the grasp force, has been relegated to a secondary plane.
Regarding this factor, the current myoprostheses approach is to set it proportional to the amplitude of the forearm EMG signals. This representation does not fully correspond to the real relationship between force and EMG and therefore, the force exerted by the myoprostheses is only an approximation of the force a sound limb would produce. Recently, commercial myoprosthesis have started to address this issue by implementing adaptive strategies for the control of the force. However, these strategies aim at preventing the objects from slipping, not actually at reproducing the behavior of a sound limb by decoding the grasp force from the EMG.
Several studies, following a number of methodologies, have been performed to estimate the force from EMG signals, achieving notable accuracies. Nevertheless, all of them focus either on a continuous estimation of the force, or on estimating it when both the force and the EMG signals have reached a stable level. These approaches are difficult to implement in a real-time system, particularly a myoprosthesis, which must comply with strict processing times in order for the action to be felt as natural by the user.
To improve the myoprosthesis response times, research has recently taken advantage of the preplanned nature of grasping movements in humans. This means that the information about the movement should be available before the movement itself. To seize this information, research has recently decanted from targeting the stable phase of the EMG (steady state) to target the transient phase for movement classification. The transient phase, i.e. from the start of the contraction until the signal reaches a stable level, has the advantage of allowing a faster response from the controller. Moreover, due to its temporal structure, the transient hints to be informative about the movement. However, this information has not been exploited for force estimation.
The objective of this dissertation is “the investigation of the grasp force from surface EMG during grasping and manipulation”. This objective was achieved in two steps. First we proved with offline HD-EMG recordings from able-bodied participants that the transient EMG contains information about the target grasp force. The last one referring to the stable plateau value of the force exerted by the subject during the movement. In order to achieve the best estimation possible, several tests were made to determine the fittest algorithm and parameters to do so. By means of this approach we made possible a faster estimation of the movement’s force of intent than with traditional continuous methods.
Finally, we integrated the methodology, algorithm and parameters found in the first stage, to develop a platform to determine if the information contained on the transient about the grasp force could be extracted in a real-time system. This platform was tested both with able-bodied participants and amputees. With that system, we were able to estimate the final GF from the transient EMG with high accuracy. On top of that, the results from the residual limb of the amputees were equivalent to the ones of the non-amputees. This paves the way towards controllers of myoprostheses that are faster and more akin to human behavior.
We also contributed to the acquisition and publication of another database of HD-EMG recordings and hand kinematics. This database could be used to test not only the classic methods of control but also the extraction of the neural drive (high sample frequency of the EMG), performing electrode selection (HD-EMG data) and the development of proportional methods of control (hand kinematics data).
Overall, we increased our knowledge about the physiological relationship under investigation: EMG and force, a rapport that is often oversimplified in the myoprostheses field. Our approach of estimating the grasp force from the transient EMG sheds new light on the relationship of both elements and on the information contained on the transient. Finally, our methodology indicates a faster way to obtain an estimate of the force of intent that can be exploited to develop robust and natural strategies of control for myoprostheses.
File
| Nome file | Dimensione |
|---|---|
| PhDTesis...gada2.pdf | 27.59 Mb |
Contatta l'autore |
|