logo SBA

ETD

Archivio digitale delle tesi discusse presso l’Università di Pisa

Tesi etd-12122022-113122


Tipo di tesi
Tesi di dottorato di ricerca
Autore
TURCO, ENRICO
URN
etd-12122022-113122
Titolo
Soft manipulation with embedded and environmental constraints
Settore scientifico disciplinare
ING-INF/04
Corso di studi
SMART INDUSTRY
Relatori
tutor Prof. Prattichizzo, Domenico
Parole chiave
  • environmental constraints exploitation
  • grasp planning
  • grasping
  • robotics
  • soft robotic hands
Data inizio appello
16/12/2022
Consultabilità
Non consultabile
Data di rilascio
16/12/2062
Riassunto
Soft manipulation approaches constitute a significant shift in the manipulation perspective; new strategies are to facilitate robust interactions with objects, allowing to deal with uncertainties. This type of approach, which leverages the environment to help the grasp, is known as Environmental Constraint Exploitation (ECE), and it has been proven to increase grasp robustness while minimizing planning efforts. The essential item for exploiting these constraints is the softness of hands, i.e., their embodied ability to comply and adapt to the features of the environment.
The starting point to achieve stable grasps is to replicate the way humans exploit environmental constraints. In this regard, a grasping strategy implemented mimicking the behavior of the human hand is the so-called slide-to-edge grasp, in which an object is dragged towards the edge of a surface and then grasped from the side.
In this thesis, two new approaches to performing such a manipulation strategy with an anthropomorphic
soft hand are presented.
A paradigm change is required when the employed gripper presents non-anthropomorphic elements, such as the embodiment of a constraint in its structure (e.g., a scoop-like flat surface working as a second palm). This thesis shows how, in this case, the gripper design features can be leveraged to achieve robust grasps exploiting both embedded and environmental constraints. Two different approaches are adopted. The first computes the gripper pre-grasp pose solving an analytical optimization problem; the second is data-driven and based on human demonstrations. Observing how humans would use a non-anthropomorphic gripper with kinesthetic teaching allowed us to understand how to exploit embedded constraints in novel ways. The newly devised method based on Learning from Demonstration enables computing the pre-grasp poses faster and using only a few training data.
Although improvements need to be deployed to increase the performance of the presented methodologies, experimental results are promising. Indeed, the proposed grasping strategies have the potential to extend the use of robotic manipulation beyond the industry area, allowing robotic systems to operate in unstructured and previously unseen scenarios.
File