logo SBA

ETD

Archivio digitale delle tesi discusse presso l’Università di Pisa

Tesi etd-11142018-230013


Tipo di tesi
Tesi di laurea magistrale
Autore
CORDELLA, GIOVANNI MARCO
URN
etd-11142018-230013
Titolo
Continuous Control of UGV for Mapless Navigation: a Virtual-to-Real Deep Reinforcement Learning Approach
Dipartimento
INGEGNERIA DELL'INFORMAZIONE
Corso di studi
INGEGNERIA ROBOTICA E DELL'AUTOMAZIONE
Relatori
relatore Prof.ssa Pallottino, Lucia
relatore Ing. Franchi, Michele
Parole chiave
  • Actor-Critic
  • Artificial Intelligence
  • autonomus robot
  • continuous action space
  • deep reinforcement learning
  • Machine learning
Data inizio appello
10/12/2018
Consultabilità
Completa
Riassunto
The autonomous mobile robot must be able to adapt its skills in order to react adequately in complex and dynamic environments. Different approaches can be used to handle these kinds of applications: map-based navigation or map-less navigation. This work aims to prove that robot can goes towards target position, detecting and avoiding obstacles based on low dimensional data directly from sensors without the knowledge of the environment map. This goal is achieved using Deep Reinforcement Learning approach by a virtual-to-real strategy; thus, networks are trained in simulations before being inferred in real world. DRL method is Actor-Critic method and the training process is the Deterministic Policy Gradient update process.
File