ETD

Archivio digitale delle tesi discusse presso l'Università di Pisa

Tesi etd-01232021-153815


Tipo di tesi
Tesi di laurea magistrale
Autore
CENCESCHI, LORENZO
URN
etd-01232021-153815
Titolo
Previous trial effect in human manipulation modeled through Iterative Learning Control
Dipartimento
INGEGNERIA DELL'INFORMAZIONE
Corso di studi
INGEGNERIA ROBOTICA E DELL'AUTOMAZIONE
Relatori
relatore Prof. Bicchi, Antonio
relatore Prof. Bianchi, Matteo
relatore Dott. Angelini, Franco
Parole chiave
  • previous trial effect
  • iterative learning control
  • modeling control and learning
Data inizio appello
25/02/2021
Consultabilità
Tesi non consultabile
Riassunto
While performing repetitive tasks, humans can exploit previous experiences to improve their
motor performance; this ability can be recognized in a wide range of actions, i.e. our capacity
of grasping and manipulating objects in uncertain conditions.
Intending to describe such behavior from a mathematical point of view, I consider ex-
periments where participants are required to lift an object with unexpected mass distribution
and maintain it vertical minimizing tilt. By repeating multiple times the same action, par-
ticipants could learn the correct motor action for a proper task accomplishment.
I propose models combining reactive terms and learned anticipatory action to explain
experimental data, featuring infra-trial and inter-trial memory, and the effect of slow and fast
adaptive sensory receptors.
Inspired by experimental humans behavior, I propose a novel sufficient condition for
convergence of SISO non-linear systems piloted by a linear iterative learning controller with
feedback and feed-forward actions. The proposed architectures and a general-purpose state
of the art model are compared for effectiveness in explaining experimental data. Proposed
algorithms outperform the state of the art in all considered validation routines. Global and
within-trial human behaviors are predicted with 88% of accuracy in nominal conditions.
Very good performance is maintained even when the object center of mass is moved. Without
any further identification, the accuracy is here maintained up to the 83%
File