ETD

Archivio digitale delle tesi discusse presso l'Università di Pisa

Tesi etd-11052021-183510


Tipo di tesi
Tesi di laurea magistrale
Autore
IUCULANO, MATILDE
URN
etd-11052021-183510
Titolo
Model-based and data-driven approaches for the temporal reconstruction of the whole muscular-skeletal state of the human upper limb from scarce sensory information. Applications to advanced human body sensing and trans-radial prosthesis control.
Dipartimento
INGEGNERIA DELL'INFORMAZIONE
Corso di studi
INGEGNERIA ROBOTICA E DELL'AUTOMAZIONE
Relatori
relatore Bianchi, Matteo
relatore Averta, Giuseppe Bruno
relatore Barontini, Federica
Parole chiave
  • kinematic model
  • Simulink
  • Arduino
  • TinyML
  • neural network
  • human upper limb
  • robotic protheses
  • control
  • electromyographic
  • extended Kalman filter
  • functional principal component analysis
  • joint angle
  • synergies
  • minimum variance estimation
  • principal component analysis
  • central nervous system
  • non-negative matrix factorization time varying
  • non-negative matrix factorization
  • inertial measurement units
Data inizio appello
25/11/2021
Consultabilità
Non consultabile
Data di rilascio
25/11/2024
Riassunto
In this work, I investigate how the synergistic coupling between different upper limb degrees of freedom can be exploited to estimate the temporal evolution of the muscular-skeletal state (electromyographic EMG and kinematics measurements) of the whole limb, from a reduced number of noisy data.

First, I applied the functional component analysis to EMG and postural data of human upper limb to obtain the basis functions. Then, I combined this functions with the scarce sensory information to solve minimum variance estimation. Furthermore, this approach integrated with an optimization problem allows us to identify which degrees of freedom should be measured to minimize the a posteriori covariance matrix.

Finally, leveraging the coupling between the upper limb degrees of freedom, I developed a control strategy for robotic trans-radial prostheses in which a neural network is trained to predict -from a given motion of the residual limb composed by the shoulder and the elbow- the most likely motion of the three-degrees of freedom wrist. I performed a first implementation of the sensing and computational framework using four Inertial Measurement Units to track the motion of the residual upper limb that was fed into the motion predictor. The deployment of the neural architecture was performed at the edge on an Arduino Nano 33 BLE board, minimizing the usage of external computational sources. Extensive experiments to test the proposed framework were carried out using a virtual avatar.
File