Objective. The journey of a bionic prosthetic user is characterized by the opportunities and limitations involved in adopting a device (the prosthesis) that should enable activities of daily living (ADL). Within this context, experiencing a bionic hand as a functional (and, possibly, embodied) limb constitutes the premise for mitigating the risk of its abandonment through the continuous use of the device. To achieve such a result, different aspects must be considered for making the artificial limb an effective support for carrying out ADLs. Among them, intuitive and robust control is fundamental to improving amputees’ quality of life using upper limb prostheses. Still, as artificial proprioception is essential to perceive the prosthesis movement without constant visual attention, a good control framework may not be enough to restore practical functionality to the limb. To overcome this, bidirectional communication between the user and the prosthesis has been recently introduced and is a requirement of utmost importance in developing prosthetic hands. Indeed, closing the control loop between the user and a prosthesis by providing artificial sensory feedback is a fundamental step towards the complete restoration of the lost sensory-motor functions. Within my PhD work, I proposed the development of a more controllable and sensitive human-like hand prosthesis, i.e., the Hannes prosthetic hand, to improve its usability and effectiveness. Approach. To achieve the objectives of this thesis work, I developed a modular and scalable software and firmware architecture to control the Hannes prosthetic multi-Degree of Freedom (DoF) system and to fit all users’ needs (hand aperture, wrist rotation, and wrist flexion in different combinations). On top of this, I developed several Pattern Recognition (PR) algorithms to translate electromyographic (EMG) activity into complex movements. However, stability and repeatability were still unmet requirements in multi-DoF upper limb systems; hence, I started by investigating different strategies to produce a more robust control. To do this, EMG signals were collected from trans-radial amputees using an array of up to six sensors placed over the skin. Secondly, I developed a vibrotactile system to implement haptic feedback to restore proprioception and create a bidirectional connection between the user and the prosthesis. Similarly, I implemented an object stiffness detection to restore tactile sensation able to connect the user with the external word. This closed-loop control between EMG and vibration feedback is essential to implementing a Bidirectional Body - Machine Interface to impact amputees’ daily life strongly. For each of these three activities: (i) implementation of robust pattern recognition control algorithms, (ii) restoration of proprioception, and (iii) restoration of the feeling of the grasped object's stiffness, I performed a study where data from healthy subjects and amputees was collected, in order to demonstrate the efficacy and usability of my implementations. In each study, I evaluated both the algorithms and the subjects’ ability to use the prosthesis by means of the F1Score parameter (offline) and the Target Achievement Control test-TAC (online). With this test, I analyzed the error rate, path efficiency, and time efficiency in completing different tasks. Main results. Among the several tested methods for Pattern Recognition, the Non-Linear Logistic Regression (NLR) resulted to be the best algorithm in terms of F1Score (99%, robustness), whereas the minimum number of electrodes needed for its functioning was determined to be 4 in the conducted offline analyses. Further, I demonstrated that its low computational burden allowed its implementation and integration on a microcontroller running at a sampling frequency of 300Hz (efficiency). Finally, the online implementation allowed the subject to simultaneously control the Hannes prosthesis DoFs, in a bioinspired and human-like way. In addition, I performed further tests with the same NLR-based control by endowing it with closed-loop proprioceptive feedback. In this scenario, the results achieved during the TAC test obtained an error rate of 15% and a path efficiency of 60% in experiments where no sources of information were available (no visual and no audio feedback). Such results demonstrated an improvement in the controllability of the system with an impact on user experience. Significance. The obtained results confirmed the hypothesis of improving robustness and efficiency of a prosthetic control thanks to of the implemented closed-loop approach. The bidirectional communication between the user and the prosthesis is capable to restore the loss of sensory functionality, with promising implications on direct translation in the clinical practice.
Novel Bidirectional Body - Machine Interface to Control Upper Limb Prosthesis
MARINELLI, ANDREA
2023-06-09
Abstract
Objective. The journey of a bionic prosthetic user is characterized by the opportunities and limitations involved in adopting a device (the prosthesis) that should enable activities of daily living (ADL). Within this context, experiencing a bionic hand as a functional (and, possibly, embodied) limb constitutes the premise for mitigating the risk of its abandonment through the continuous use of the device. To achieve such a result, different aspects must be considered for making the artificial limb an effective support for carrying out ADLs. Among them, intuitive and robust control is fundamental to improving amputees’ quality of life using upper limb prostheses. Still, as artificial proprioception is essential to perceive the prosthesis movement without constant visual attention, a good control framework may not be enough to restore practical functionality to the limb. To overcome this, bidirectional communication between the user and the prosthesis has been recently introduced and is a requirement of utmost importance in developing prosthetic hands. Indeed, closing the control loop between the user and a prosthesis by providing artificial sensory feedback is a fundamental step towards the complete restoration of the lost sensory-motor functions. Within my PhD work, I proposed the development of a more controllable and sensitive human-like hand prosthesis, i.e., the Hannes prosthetic hand, to improve its usability and effectiveness. Approach. To achieve the objectives of this thesis work, I developed a modular and scalable software and firmware architecture to control the Hannes prosthetic multi-Degree of Freedom (DoF) system and to fit all users’ needs (hand aperture, wrist rotation, and wrist flexion in different combinations). On top of this, I developed several Pattern Recognition (PR) algorithms to translate electromyographic (EMG) activity into complex movements. However, stability and repeatability were still unmet requirements in multi-DoF upper limb systems; hence, I started by investigating different strategies to produce a more robust control. To do this, EMG signals were collected from trans-radial amputees using an array of up to six sensors placed over the skin. Secondly, I developed a vibrotactile system to implement haptic feedback to restore proprioception and create a bidirectional connection between the user and the prosthesis. Similarly, I implemented an object stiffness detection to restore tactile sensation able to connect the user with the external word. This closed-loop control between EMG and vibration feedback is essential to implementing a Bidirectional Body - Machine Interface to impact amputees’ daily life strongly. For each of these three activities: (i) implementation of robust pattern recognition control algorithms, (ii) restoration of proprioception, and (iii) restoration of the feeling of the grasped object's stiffness, I performed a study where data from healthy subjects and amputees was collected, in order to demonstrate the efficacy and usability of my implementations. In each study, I evaluated both the algorithms and the subjects’ ability to use the prosthesis by means of the F1Score parameter (offline) and the Target Achievement Control test-TAC (online). With this test, I analyzed the error rate, path efficiency, and time efficiency in completing different tasks. Main results. Among the several tested methods for Pattern Recognition, the Non-Linear Logistic Regression (NLR) resulted to be the best algorithm in terms of F1Score (99%, robustness), whereas the minimum number of electrodes needed for its functioning was determined to be 4 in the conducted offline analyses. Further, I demonstrated that its low computational burden allowed its implementation and integration on a microcontroller running at a sampling frequency of 300Hz (efficiency). Finally, the online implementation allowed the subject to simultaneously control the Hannes prosthesis DoFs, in a bioinspired and human-like way. In addition, I performed further tests with the same NLR-based control by endowing it with closed-loop proprioceptive feedback. In this scenario, the results achieved during the TAC test obtained an error rate of 15% and a path efficiency of 60% in experiments where no sources of information were available (no visual and no audio feedback). Such results demonstrated an improvement in the controllability of the system with an impact on user experience. Significance. The obtained results confirmed the hypothesis of improving robustness and efficiency of a prosthetic control thanks to of the implemented closed-loop approach. The bidirectional communication between the user and the prosthesis is capable to restore the loss of sensory functionality, with promising implications on direct translation in the clinical practice.File | Dimensione | Formato | |
---|---|---|---|
phdunige_4425269.pdf
accesso aperto
Tipologia:
Tesi di dottorato
Dimensione
12.05 MB
Formato
Adobe PDF
|
12.05 MB | Adobe PDF | Visualizza/Apri |
I documenti in IRIS sono protetti da copyright e tutti i diritti sono riservati, salvo diversa indicazione.