Il full-text non è disponibile per scelta dell'autore.
(
Contatta l'autore)
Abstract
The number of autonomous vehicles integrated into modern transportation systems is growing on a daily basis. This introduced novel challenging tasks for the Automation and Robotics communities, such as environment perception, position estimation, path planning and motion control. Such challenging scenarios required the development of specialized control algorithms: an example is the model predictive control (MPC) algorithm, a control technique which gained more and more popularity in recent years thanks to its ability to include constraints. However, MPC strongly relies on accurate models of the system which are often hard to obtain. In recent years, thanks to the widespread of embedded systems and the development of powerful computing units, data-driven techniques have been introduced in the world of autonomous driving leveraging measured data to identify and understand complex system dynamics.
In this thesis, autonomous driving of scaled vehicles is explored from a three-fold perspective. First, a control architecture tailored for controlling a Jetracer autonomous racecar is proposed. This platform is a 1:14 model of a self-driving vehicle and represents a consolidated platform for implementing and testing various control algorithms, facilitating experimentation and validation of autonomous navigation strategies. Second, given the absence of a perfect model for the racecar, data-driven MPC techniques are investigated: through simulations on both linear and nonlinear systems, insights are highlighted about the performance and challenges of the application of data-driven MPC formulations on real-world scenarios where noisy measurements or inaccuracies in model parameters could be present. Third, the control architecture is experimentally validated on the Jetracer platform through the deployment of both PID controller and data-driven MPC scheme demonstrating the practical capabilities and modularity of the developed software infrastructure in real-world applications.
Abstract
The number of autonomous vehicles integrated into modern transportation systems is growing on a daily basis. This introduced novel challenging tasks for the Automation and Robotics communities, such as environment perception, position estimation, path planning and motion control. Such challenging scenarios required the development of specialized control algorithms: an example is the model predictive control (MPC) algorithm, a control technique which gained more and more popularity in recent years thanks to its ability to include constraints. However, MPC strongly relies on accurate models of the system which are often hard to obtain. In recent years, thanks to the widespread of embedded systems and the development of powerful computing units, data-driven techniques have been introduced in the world of autonomous driving leveraging measured data to identify and understand complex system dynamics.
In this thesis, autonomous driving of scaled vehicles is explored from a three-fold perspective. First, a control architecture tailored for controlling a Jetracer autonomous racecar is proposed. This platform is a 1:14 model of a self-driving vehicle and represents a consolidated platform for implementing and testing various control algorithms, facilitating experimentation and validation of autonomous navigation strategies. Second, given the absence of a perfect model for the racecar, data-driven MPC techniques are investigated: through simulations on both linear and nonlinear systems, insights are highlighted about the performance and challenges of the application of data-driven MPC formulations on real-world scenarios where noisy measurements or inaccuracies in model parameters could be present. Third, the control architecture is experimentally validated on the Jetracer platform through the deployment of both PID controller and data-driven MPC scheme demonstrating the practical capabilities and modularity of the developed software infrastructure in real-world applications.
Tipologia del documento
Tesi di laurea
(Laurea magistrale)
Autore della tesi
Drudi, Andrea
Relatore della tesi
Correlatore della tesi
Scuola
Corso di studio
Ordinamento Cds
DM270
Parole chiave
data-driven MPC,Jetracer,ROS2
Data di discussione della Tesi
18 Marzo 2024
URI
Altri metadati
Tipologia del documento
Tesi di laurea
(NON SPECIFICATO)
Autore della tesi
Drudi, Andrea
Relatore della tesi
Correlatore della tesi
Scuola
Corso di studio
Ordinamento Cds
DM270
Parole chiave
data-driven MPC,Jetracer,ROS2
Data di discussione della Tesi
18 Marzo 2024
URI
Gestione del documento: