University of Patras
Cooperative graph signal processing and learning for situational awareness of autonomous vehicles
Abstract
dc:descriptionThe aim of this dissertation was to study and develop innovative and efficient cooperative multi-modal data fusion solutions for cooperative localization and tracking of connected and automated vehicles, implemented in a distributed manner by an ego vehicle along with either its connected neighborhood or non-cooperative static landmarks. Critical point of the methodology was the treatment of the network of connected and automated vehicles as an undirected graph, where its nodes represent vehicles while its edges the connectivity links. Based on this approach, the Graph Laplacian Processing concept, a graph signal processing tool, has been devised, together with traditional estimation and learning algorithms like least-squares minimization, Kalman filter, Extended Kalman Filter, information diffusion on graphs, etc., in order to realize the goal for effective cooperative multi-modal fusion, and in general, cooperative localization and tracking. Key aspect of Graph Laplacian Processing is that it transforms the non-linear relative vehicular measurements into linear ones via differential coordinates, thus leading to a general linear system. Furthermore, data driven deep learning solutions have been exploited for specific parts of traditional optimization algorithms, in order to efficiently track the underlying system uncertainty, and as a matter of fact improving the cooperative multi-modal fusion operation. The focus was on designing fully distributed solutions, where each vehicle interacts only with its direct neighborhood, due to scalability, adaptability, synchronization, etc., thus avoiding a costly centralized fusion center which monitors the entire vehicular network. Following that, vehicles will be capable of estimating not only local but also global location parameters, related to their own 1-hop neighborhood, as well as to all other vehicles belonging to the same network or cluster. As such, sensing ability can now be extended beyond the range of an individual vehicle, thus enabling vehicles to be fully aware about the state of vehicles even kilometers away. Experimental setup included traffic data from many vehicles extracted by renowned autonomous driving simulators, as well as network data in order to realistically model the vehicular communication conditions. In addition to that, available public datasets including parameters from real vehicles moving in a city, as well as data collected by a real operating vehicle, have also been used in the experiments. Evaluation study considered the impact of various critical scenarios, e.g., size of neighborhood, noise of multi-modal measurements, network delay, convergence ability, execution time, etc., in order to demonstrate the benefits of this dissertation's proposed approaches.
Degree
thesis:*- Grantor dc:publisher
- University of Patras
- Year dc:date
- 2024
Author and committee
dc:creator, dc:contributor.*- Authors dc:creator
-
- Piperigkos, Nikolaos-Antonios
- Πιπερίγκος, Νικόλαος-Αντώνιος
Subjects
dc:subject × 16- Εντοπισμός θέσης
- Λαπλασιανή επεξεργασία γράφου
- Φίλτρο Καλμαν
- Σύντηξη αισθητήρων
- Διασυνδεδεμένα και αυτόνομα οχήματα
- Localization
- Graph laplacian processing
- Kalman filter
- Sensor fusion
- Connected and automated vehicles
- Επιστήμη Ηλεκτρολόγου Μηχανικού, Ηλεκτρονικού Μηχανικού, Μηχανικού Η/Υ
- Επιστήμες Μηχανικού και Τεχνολογία
- Αυτοματισμοί και Συστήματα ελέγχου
- Electrical Engineering, Electronic Engineering, Information Engineering
- Engineering and Technology
- Automation and control systems
Rights
- Language dc:language
- eng
Identifiers
dc:identifier.*- Identifier
- 10.12681/eadd/57542