With the increasing interest in autonomous vehicles, a growing number of applications require the use of robotic platforms in urban areas. Examples range from surveillance drones, guide and wayfinding robots, autonomous street sweepers, and parcel delivery robots. The latter in particular have attracted a lot of attention, with several vehicles being developed and experimented with. The promise is for these small, electrically powered, and autonomous vehicles to reduce the congestion and environmental impact of deliveries, all while decreasing costs for the operators. Indeed, autonomous delivery robots could navigate on sidewalks and in pedestrian areas, replacing large and inefficient delivery vans. The task of autonomously navigating in pedestrian areas, however, is still a demanding problem, and an open one. The challenges of such a task are many, deriving from the unstructured nature of an environment built solely for humans and ranging from the requirement to handle locations with varying layouts (tight sidewalks, vast squares, unpaved park paths...), to the lack of strict norms governing navigation, to the need to interact at close quarters with pedestrians. This dissertation presents algorithmic solutions for different components of an autonomous driving stack specifically designed to handle the challenges of pedestrian areas. First, a perception and mapping algorithm is developed to detect and map dangerous locations automatically. The resulting map can be employed as a global planning map or to supplement online perception algorithms in challenging situations. As a second task, the localization problem is treated in depth. A detailed analysis of state-of-the-art map-based localization algorithms is performed, while also developing a robust localization architecture. Finally, a socially aware local planner is developed, enabling robotic vehicles to efficiently and safely interact with pedestrians in collision avoidance maneuvers. As a final contribution of this work, a robotic guide for Blind or Visually Impaired people is conceived, developed, and validated, gathering feedback from potential users. All algorithms are validated experimentally on a two-wheeled actively stabilized robot.
L’interesse verso veicoli autonomi con le più svariate applicazioni è in costante ascesa Esempi di utilizzo spaziano da droni per la sorveglianza, robot impiegati come guide sia turistiche che sanitarie (per ipovedenti, ad esempio), fino a veicoli autonomi per la pulizia delle strade o robot per la consegna di corrispondenza. Questi ultimi in particolare hanno suscitato un notevole interesse nelle aziende di logistica, con diversi veicoli in fase di sviluppo e sperimentazione. La speranza è che questi veicoli piccoli, elettrici e autonomi possano ridurre la congestione e l'impatto ambientale delle consegne, diminuendo al contempo i costi per gli operatori. Veicoli autonomi per la consegna potrebbero, infatti, navigare su marciapiedi e nelle aree pedonali, sostituendo i grandi e inefficienti furgoni utilizzati oggi. Navigare autonomamente in aree pedonali è tuttavia un problema non banale e senza una soluzione universalmente accettata. Le sfide dell’ambiente urbano sono molteplici, derivanti dalla natura non strutturata di un ambiente costruito esclusivamente per gli esseri umani e che vanno dalla necessità di gestire luoghi con layout variabili (marciapiedi stretti, piazze ampie, sentieri non asfaltati nei parchi...), la mancanza di norme rigorose che regolamentino la navigazione, fino alla necessità di interagire con pedoni a distanza ravvicinata. Questa tesi presenta diversi algoritmi sviluppati per una piattaforma di guida autonoma appositamente progettata per affrontare le sfide delle aree pedonali. Innanzitutto, viene sviluppato un algoritmo di percezione e mappatura per rilevare e mappare automaticamente le aree non navigabili. La mappa risultante può essere impiegata come mappa di pianificazione globale o per integrare algoritmi di percezione online in situazioni difficili. In secondo luogo, viene trattato in dettaglio il problema della localizzazione. Viene eseguita un'analisi dettagliata degli algoritmi di localizzazione basati su mappa allo stato dell'arte, sviluppando nel contempo un'architettura di localizzazione robusta. Infine, viene sviluppato un pianificatore locale “socially aware”, che consente ai veicoli robotici di interagire in modo efficiente e sicuro con i pedoni nelle manovre di evitamento delle collisioni. Come contributo finale di questo lavoro, viene concepito, sviluppato e validato un robot guida per persone non vedenti o ipovedenti, raccogliendo feedback dai potenziali utenti. Tutti gli algoritmi sono validati sperimentalmente su un robot a due ruote con struttura a pendolo inverso.
Navigation algorithms for autonomous robots in pedestrian areas
Mozzarelli, Luca
2023/2024
Abstract
With the increasing interest in autonomous vehicles, a growing number of applications require the use of robotic platforms in urban areas. Examples range from surveillance drones, guide and wayfinding robots, autonomous street sweepers, and parcel delivery robots. The latter in particular have attracted a lot of attention, with several vehicles being developed and experimented with. The promise is for these small, electrically powered, and autonomous vehicles to reduce the congestion and environmental impact of deliveries, all while decreasing costs for the operators. Indeed, autonomous delivery robots could navigate on sidewalks and in pedestrian areas, replacing large and inefficient delivery vans. The task of autonomously navigating in pedestrian areas, however, is still a demanding problem, and an open one. The challenges of such a task are many, deriving from the unstructured nature of an environment built solely for humans and ranging from the requirement to handle locations with varying layouts (tight sidewalks, vast squares, unpaved park paths...), to the lack of strict norms governing navigation, to the need to interact at close quarters with pedestrians. This dissertation presents algorithmic solutions for different components of an autonomous driving stack specifically designed to handle the challenges of pedestrian areas. First, a perception and mapping algorithm is developed to detect and map dangerous locations automatically. The resulting map can be employed as a global planning map or to supplement online perception algorithms in challenging situations. As a second task, the localization problem is treated in depth. A detailed analysis of state-of-the-art map-based localization algorithms is performed, while also developing a robust localization architecture. Finally, a socially aware local planner is developed, enabling robotic vehicles to efficiently and safely interact with pedestrians in collision avoidance maneuvers. As a final contribution of this work, a robotic guide for Blind or Visually Impaired people is conceived, developed, and validated, gathering feedback from potential users. All algorithms are validated experimentally on a two-wheeled actively stabilized robot.File | Dimensione | Formato | |
---|---|---|---|
thesis.pdf
non accessibile
Descrizione: main thesis file
Dimensione
121.49 MB
Formato
Adobe PDF
|
121.49 MB | Adobe PDF | Visualizza/Apri |
I documenti in POLITesi sono protetti da copyright e tutti i diritti sono riservati, salvo diversa indicazione.
https://hdl.handle.net/10589/222383