Alberto Carpineto
Design and Robotic Implementation of a Hierarchical Reinforcement Learning Control Strategy for Simultaneous Multi-Object Grasping.
Rel. Alessandro Rizzo, Egidio Falotico. Politecnico di Torino, Corso di laurea magistrale in Mechatronic Engineering (Ingegneria Meccatronica), 2026
|
|
PDF (Tesi_di_laurea)
- Tesi
Accesso limitato a: Solo utenti staff fino al 24 Gennaio 2028 (data di embargo). Licenza: Creative Commons Attribution Non-commercial No Derivatives. Download (56MB) |
Abstract
Grasping is one of the most fundamental and demanding capabilities in physical interaction: it requires the seamless coordination of perception, cognition, and motor execution. Replicating this ability in robotic systems is still an open challenge. Despite the remarkable mechanical precision of modern robotic manipulators, the vast majority of existing systems are designed to grasp one object at a time, a constraint that significantly limits their effectiveness in practical scenarios such as bin-picking, warehouse automation, and workspace clearing. This thesis addresses that limitation by designing and implementing a complete framework for Simultaneous Multi-Object Grasping (SMOG). At the heart of the approach is a novel paradigm that leverages a Hierarchical Reinforcement Learning (HRL) algorithm to analyze the scene and group multiple objects into spatial clusters, each collectable within a single grasp action.
Compared to state-of the-art graph-based methods, the proposed strategy offers two key advantages: improved recovery from grasp failures, and the ability to dynamically adapt the grasping force based on the number and spatial distribution of target objects
Relatori
Anno Accademico
Tipo di pubblicazione
Numero di pagine
Corso di laurea
Classe di laurea
Aziende collaboratrici
URI
![]() |
Modifica (riservato agli operatori) |
