ETD

Archivio digitale delle tesi discusse presso l'Università di Pisa

Tesi etd-11012016-193052


Tipo di tesi
Tesi di laurea magistrale
Autore
ACCOGLI, ALESSANDRO
URN
etd-11012016-193052
Titolo
Intention Detection EMG-based Algorithms for Upper-Limb Assistive Exoskeleton Control
Dipartimento
INGEGNERIA DELL'INFORMAZIONE
Corso di studi
INGEGNERIA BIOMEDICA
Relatori
relatore Prof. Dario, Paolo
relatore Prof. Micera, Silvestro
tutor Dott. Panarese, Alessandro
Parole chiave
  • SVM
  • onset detection
  • human machine interface
  • GMM
  • direction decoding
  • upper limb exoskeleton
Data inizio appello
02/12/2016
Consultabilità
Tesi non consultabile
Data di rilascio
02/12/2086
Riassunto
Assistive robotic technology aims at providing help to disabled individuals in order for them to perform normal activities of daily living as independently as possible. Together with new robotic devices, which can compensate for lost functions of disabled users, new control paradigms are needed to improve their ability to use such external devices naturally. This thesis project paid specific attention to the design of a Human-Machine Interface (HMI) for an exoskeleton, i.e. a wearable upper-limb assistive robot, based on the decoding of user’s intentions from non-invasive surface EMG signals. The main purpose of this work was the design and development of advanced Machine Learning (ML) algorithms to determine both the intention of reaching movements and the direction of movement, as fast and accurately as possible. An unsupervised learning process was proposed based on a sequential Gaussian Mixture Model (GMM) method to detect the user’s motion intention. Still, the computational requirements had to be minimized for real-time application. By using the compound EMG signal (a unique signal fusing all the EMG signals together), it was possible to maintain (or even improve) performances while reducing the computational load. A subject-independent set of features avoided a subject-specific selection of features, a very time consuming procedure for training the algorithm. The role of each of the seven muscles recorded from the shoulder, the arm and the forearm was analyzed to understand whether each muscle could have been excluded from the analysis to reduce the user’s discomfort and the computational load. Finally, to test and validate the GMM also online, the code was converted into a shared library written in C language and integrated in LabVIEW to be used by the control system of the exoskeleton to communicate with the device. The classification problem of reaching movement direction was solved with Support Vector Machines (SVM), a well-known ML algorithm. The classification performances were significantly improved by adding kinematic information from the exoskeleton as input to the classifier. Future works will be devoted to study the real-time application of the presented algorithms for both movement onset detection and direction decoding in people with severe arm disabilities.
File