Current developments of myoelectric upper limb prostheses are increasingly oriented toward advanced, technologically sophisticated devices aimed at improving the quality of life of users. These complex systems often incorporate artificial intelligence to manage both control and tactile feedback. At Rehab Technologies of the Istituto Italiano di Tecnologia, efforts are directed toward the development of a highly dexterous, multi-articulated bionic hand, conceived as a robotic platform for testing advanced control algorithms. A key aspect of this work concerns its mechatronic characterization, encompassing the definition of the mechanical, electronic, and control subsystems required to achieve the desired dexterity. In this thesis, several test benches were designed and evaluated, beginning with an analysis of the embedded motors and proceeding to the mathematical formulation of synergies of the joints of the individual fingers. Controlling such a complex system requires robust classification of upper-limb movements through muscular activations, with particular attention to independent movements of the joints. For this purpose, deep learning models were trained and evaluated on 16 classes of wrist and finger joints movements, and compared with a shallow machine learning approach to assess their effectiveness. The mechatronic study, thanks to the implementation of low and middle level control, led to the identification of suitable motors and to the mathematical definition of finger synergies, ensuring reliable finger actuation. Concurrently, gesture recognition based on learning algorithms applied to high density surface electromyography signals yielded preliminary results supporting the integration of neural models into embedded control systems of the bionic hand.
Gli sviluppi attuali delle protesi mioelettriche per l’arto superiore sono sempre più orientati verso dispositivi avanzati e tecnologicamente sofisticati, mirati a migliorare la qualità della vita degli utenti. Questi sistemi complessi integrano spesso tecniche di intelligenza artificiale sia per la gestione del controllo sia per la generazione del feedback tattile. Presso il laboratorio Rehab Technologies dell’Istituto Italiano di Tecnologia, gli sforzi sono focalizzati sullo sviluppo di una mano bionica altamente destrorsa e multi-articolata, concepita come piattaforma robotica per il testing di algoritmi di controllo avanzati. Un aspetto chiave di questo lavoro riguarda la sua caratterizzazione meccatronica, che comprende la definizione dei sottosistemi meccanici, elettronici e di controllo necessari per ottenere la destrezza desiderata. In questa tesi sono stati progettati e valutati diversi banchi prova, a partire dall’analisi dei motori integrati fino ad arrivare alla formulazione matematica delle sinergie articolari delle singole dita. Il controllo di un sistema così complesso richiede una classificazione robusta dei movimenti dell’arto superiore attraverso l’attivazione muscolare, con particolare attenzione ai movimenti indipendenti delle articolazioni. A tal fine, modelli di deep learning sono stati addestrati e valutati su 16 classi di movimenti di polso e dita, e confrontati con un approccio di machine learning tradizionale per valutarne l’efficacia. Lo studio meccatronico, grazie all’implementazione di controlli a basso e medio livello, ha consentito l’identificazione dei motori più adeguati e la definizione matematica delle sinergie digitali, garantendo un’azione delle dita affidabile. Parallelamente, il riconoscimento di gesti basato su algoritmi di apprendimento applicati a segnali di elettromiografia di superficie ad alta densità ha prodotto risultati preliminari che supportano l’integrazione di modelli neurali nei sistemi di controllo embedded della mano bionica.
Integrated test bench for the mechatronic characterization and control of a dexterous bionic hand
BENINCA', GIULIO
2024/2025
Abstract
Current developments of myoelectric upper limb prostheses are increasingly oriented toward advanced, technologically sophisticated devices aimed at improving the quality of life of users. These complex systems often incorporate artificial intelligence to manage both control and tactile feedback. At Rehab Technologies of the Istituto Italiano di Tecnologia, efforts are directed toward the development of a highly dexterous, multi-articulated bionic hand, conceived as a robotic platform for testing advanced control algorithms. A key aspect of this work concerns its mechatronic characterization, encompassing the definition of the mechanical, electronic, and control subsystems required to achieve the desired dexterity. In this thesis, several test benches were designed and evaluated, beginning with an analysis of the embedded motors and proceeding to the mathematical formulation of synergies of the joints of the individual fingers. Controlling such a complex system requires robust classification of upper-limb movements through muscular activations, with particular attention to independent movements of the joints. For this purpose, deep learning models were trained and evaluated on 16 classes of wrist and finger joints movements, and compared with a shallow machine learning approach to assess their effectiveness. The mechatronic study, thanks to the implementation of low and middle level control, led to the identification of suitable motors and to the mathematical definition of finger synergies, ensuring reliable finger actuation. Concurrently, gesture recognition based on learning algorithms applied to high density surface electromyography signals yielded preliminary results supporting the integration of neural models into embedded control systems of the bionic hand.| File | Dimensione | Formato | |
|---|---|---|---|
|
Beninca_Giulio.pdf
embargo fino al 05/06/2027
Dimensione
6.59 MB
Formato
Adobe PDF
|
6.59 MB | Adobe PDF |
The text of this website © Università degli studi di Padova. Full Text are published under a non-exclusive license. Metadata are under a CC0 License
https://hdl.handle.net/20.500.12608/99590